1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2005 - 2016 Broadcom 4 * All rights reserved. 5 * 6 * Contact Information: 7 * linux-drivers@emulex.com 8 * 9 * Emulex 10 * 3333 Susan Street 11 * Costa Mesa, CA 92626 12 */ 13 14 #include <linux/prefetch.h> 15 #include <linux/module.h> 16 #include "be.h" 17 #include "be_cmds.h" 18 #include <asm/div64.h> 19 #include <linux/if_bridge.h> 20 #include <net/busy_poll.h> 21 #include <net/vxlan.h> 22 23 MODULE_DESCRIPTION(DRV_DESC); 24 MODULE_AUTHOR("Emulex Corporation"); 25 MODULE_LICENSE("GPL"); 26 27 /* num_vfs module param is obsolete. 28 * Use sysfs method to enable/disable VFs. 29 */ 30 static unsigned int num_vfs; 31 module_param(num_vfs, uint, 0444); 32 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize"); 33 34 static ushort rx_frag_size = 2048; 35 module_param(rx_frag_size, ushort, 0444); 36 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data."); 37 38 /* Per-module error detection/recovery workq shared across all functions. 39 * Each function schedules its own work request on this shared workq. 40 */ 41 static struct workqueue_struct *be_err_recovery_workq; 42 43 static const struct pci_device_id be_dev_ids[] = { 44 #ifdef CONFIG_BE2NET_BE2 45 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) }, 46 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) }, 47 #endif /* CONFIG_BE2NET_BE2 */ 48 #ifdef CONFIG_BE2NET_BE3 49 { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) }, 50 { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) }, 51 #endif /* CONFIG_BE2NET_BE3 */ 52 #ifdef CONFIG_BE2NET_LANCER 53 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)}, 54 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)}, 55 #endif /* CONFIG_BE2NET_LANCER */ 56 #ifdef CONFIG_BE2NET_SKYHAWK 57 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)}, 58 { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)}, 59 #endif /* CONFIG_BE2NET_SKYHAWK */ 60 { 0 } 61 }; 62 MODULE_DEVICE_TABLE(pci, be_dev_ids); 63 64 /* Workqueue used by all functions for deferring cmd calls to the adapter */ 65 static struct workqueue_struct *be_wq; 66 67 /* UE Status Low CSR */ 68 static const char * const ue_status_low_desc[] = { 69 "CEV", 70 "CTX", 71 "DBUF", 72 "ERX", 73 "Host", 74 "MPU", 75 "NDMA", 76 "PTC ", 77 "RDMA ", 78 "RXF ", 79 "RXIPS ", 80 "RXULP0 ", 81 "RXULP1 ", 82 "RXULP2 ", 83 "TIM ", 84 "TPOST ", 85 "TPRE ", 86 "TXIPS ", 87 "TXULP0 ", 88 "TXULP1 ", 89 "UC ", 90 "WDMA ", 91 "TXULP2 ", 92 "HOST1 ", 93 "P0_OB_LINK ", 94 "P1_OB_LINK ", 95 "HOST_GPIO ", 96 "MBOX ", 97 "ERX2 ", 98 "SPARE ", 99 "JTAG ", 100 "MPU_INTPEND " 101 }; 102 103 /* UE Status High CSR */ 104 static const char * const ue_status_hi_desc[] = { 105 "LPCMEMHOST", 106 "MGMT_MAC", 107 "PCS0ONLINE", 108 "MPU_IRAM", 109 "PCS1ONLINE", 110 "PCTL0", 111 "PCTL1", 112 "PMEM", 113 "RR", 114 "TXPB", 115 "RXPP", 116 "XAUI", 117 "TXP", 118 "ARM", 119 "IPC", 120 "HOST2", 121 "HOST3", 122 "HOST4", 123 "HOST5", 124 "HOST6", 125 "HOST7", 126 "ECRC", 127 "Poison TLP", 128 "NETC", 129 "PERIPH", 130 "LLTXULP", 131 "D2P", 132 "RCON", 133 "LDMA", 134 "LLTXP", 135 "LLTXPB", 136 "Unknown" 137 }; 138 139 #define BE_VF_IF_EN_FLAGS (BE_IF_FLAGS_UNTAGGED | \ 140 BE_IF_FLAGS_BROADCAST | \ 141 BE_IF_FLAGS_MULTICAST | \ 142 BE_IF_FLAGS_PASS_L3L4_ERRORS) 143 144 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q) 145 { 146 struct be_dma_mem *mem = &q->dma_mem; 147 148 if (mem->va) { 149 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va, 150 mem->dma); 151 mem->va = NULL; 152 } 153 } 154 155 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q, 156 u16 len, u16 entry_size) 157 { 158 struct be_dma_mem *mem = &q->dma_mem; 159 160 memset(q, 0, sizeof(*q)); 161 q->len = len; 162 q->entry_size = entry_size; 163 mem->size = len * entry_size; 164 mem->va = dma_alloc_coherent(&adapter->pdev->dev, mem->size, 165 &mem->dma, GFP_KERNEL); 166 if (!mem->va) 167 return -ENOMEM; 168 return 0; 169 } 170 171 static void be_reg_intr_set(struct be_adapter *adapter, bool enable) 172 { 173 u32 reg, enabled; 174 175 pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, 176 ®); 177 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK; 178 179 if (!enabled && enable) 180 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK; 181 else if (enabled && !enable) 182 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK; 183 else 184 return; 185 186 pci_write_config_dword(adapter->pdev, 187 PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg); 188 } 189 190 static void be_intr_set(struct be_adapter *adapter, bool enable) 191 { 192 int status = 0; 193 194 /* On lancer interrupts can't be controlled via this register */ 195 if (lancer_chip(adapter)) 196 return; 197 198 if (be_check_error(adapter, BE_ERROR_EEH)) 199 return; 200 201 status = be_cmd_intr_set(adapter, enable); 202 if (status) 203 be_reg_intr_set(adapter, enable); 204 } 205 206 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted) 207 { 208 u32 val = 0; 209 210 if (be_check_error(adapter, BE_ERROR_HW)) 211 return; 212 213 val |= qid & DB_RQ_RING_ID_MASK; 214 val |= posted << DB_RQ_NUM_POSTED_SHIFT; 215 216 wmb(); 217 iowrite32(val, adapter->db + DB_RQ_OFFSET); 218 } 219 220 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo, 221 u16 posted) 222 { 223 u32 val = 0; 224 225 if (be_check_error(adapter, BE_ERROR_HW)) 226 return; 227 228 val |= txo->q.id & DB_TXULP_RING_ID_MASK; 229 val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT; 230 231 wmb(); 232 iowrite32(val, adapter->db + txo->db_offset); 233 } 234 235 static void be_eq_notify(struct be_adapter *adapter, u16 qid, 236 bool arm, bool clear_int, u16 num_popped, 237 u32 eq_delay_mult_enc) 238 { 239 u32 val = 0; 240 241 val |= qid & DB_EQ_RING_ID_MASK; 242 val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT); 243 244 if (be_check_error(adapter, BE_ERROR_HW)) 245 return; 246 247 if (arm) 248 val |= 1 << DB_EQ_REARM_SHIFT; 249 if (clear_int) 250 val |= 1 << DB_EQ_CLR_SHIFT; 251 val |= 1 << DB_EQ_EVNT_SHIFT; 252 val |= num_popped << DB_EQ_NUM_POPPED_SHIFT; 253 val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT; 254 iowrite32(val, adapter->db + DB_EQ_OFFSET); 255 } 256 257 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped) 258 { 259 u32 val = 0; 260 261 val |= qid & DB_CQ_RING_ID_MASK; 262 val |= ((qid & DB_CQ_RING_ID_EXT_MASK) << 263 DB_CQ_RING_ID_EXT_MASK_SHIFT); 264 265 if (be_check_error(adapter, BE_ERROR_HW)) 266 return; 267 268 if (arm) 269 val |= 1 << DB_CQ_REARM_SHIFT; 270 val |= num_popped << DB_CQ_NUM_POPPED_SHIFT; 271 iowrite32(val, adapter->db + DB_CQ_OFFSET); 272 } 273 274 static int be_dev_mac_add(struct be_adapter *adapter, const u8 *mac) 275 { 276 int i; 277 278 /* Check if mac has already been added as part of uc-list */ 279 for (i = 0; i < adapter->uc_macs; i++) { 280 if (ether_addr_equal(adapter->uc_list[i].mac, mac)) { 281 /* mac already added, skip addition */ 282 adapter->pmac_id[0] = adapter->pmac_id[i + 1]; 283 return 0; 284 } 285 } 286 287 return be_cmd_pmac_add(adapter, mac, adapter->if_handle, 288 &adapter->pmac_id[0], 0); 289 } 290 291 static void be_dev_mac_del(struct be_adapter *adapter, int pmac_id) 292 { 293 int i; 294 295 /* Skip deletion if the programmed mac is 296 * being used in uc-list 297 */ 298 for (i = 0; i < adapter->uc_macs; i++) { 299 if (adapter->pmac_id[i + 1] == pmac_id) 300 return; 301 } 302 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0); 303 } 304 305 static int be_mac_addr_set(struct net_device *netdev, void *p) 306 { 307 struct be_adapter *adapter = netdev_priv(netdev); 308 struct device *dev = &adapter->pdev->dev; 309 struct sockaddr *addr = p; 310 int status; 311 u8 mac[ETH_ALEN]; 312 u32 old_pmac_id = adapter->pmac_id[0]; 313 314 if (!is_valid_ether_addr(addr->sa_data)) 315 return -EADDRNOTAVAIL; 316 317 /* Proceed further only if, User provided MAC is different 318 * from active MAC 319 */ 320 if (ether_addr_equal(addr->sa_data, adapter->dev_mac)) 321 return 0; 322 323 /* BE3 VFs without FILTMGMT privilege are not allowed to set its MAC 324 * address 325 */ 326 if (BEx_chip(adapter) && be_virtfn(adapter) && 327 !check_privilege(adapter, BE_PRIV_FILTMGMT)) 328 return -EPERM; 329 330 /* if device is not running, copy MAC to netdev->dev_addr */ 331 if (!netif_running(netdev)) 332 goto done; 333 334 /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT 335 * privilege or if PF did not provision the new MAC address. 336 * On BE3, this cmd will always fail if the VF doesn't have the 337 * FILTMGMT privilege. This failure is OK, only if the PF programmed 338 * the MAC for the VF. 339 */ 340 mutex_lock(&adapter->rx_filter_lock); 341 status = be_dev_mac_add(adapter, (u8 *)addr->sa_data); 342 if (!status) { 343 344 /* Delete the old programmed MAC. This call may fail if the 345 * old MAC was already deleted by the PF driver. 346 */ 347 if (adapter->pmac_id[0] != old_pmac_id) 348 be_dev_mac_del(adapter, old_pmac_id); 349 } 350 351 mutex_unlock(&adapter->rx_filter_lock); 352 /* Decide if the new MAC is successfully activated only after 353 * querying the FW 354 */ 355 status = be_cmd_get_active_mac(adapter, adapter->pmac_id[0], mac, 356 adapter->if_handle, true, 0); 357 if (status) 358 goto err; 359 360 /* The MAC change did not happen, either due to lack of privilege 361 * or PF didn't pre-provision. 362 */ 363 if (!ether_addr_equal(addr->sa_data, mac)) { 364 status = -EPERM; 365 goto err; 366 } 367 368 /* Remember currently programmed MAC */ 369 ether_addr_copy(adapter->dev_mac, addr->sa_data); 370 done: 371 eth_hw_addr_set(netdev, addr->sa_data); 372 dev_info(dev, "MAC address changed to %pM\n", addr->sa_data); 373 return 0; 374 err: 375 dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data); 376 return status; 377 } 378 379 /* BE2 supports only v0 cmd */ 380 static void *hw_stats_from_cmd(struct be_adapter *adapter) 381 { 382 if (BE2_chip(adapter)) { 383 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va; 384 385 return &cmd->hw_stats; 386 } else if (BE3_chip(adapter)) { 387 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va; 388 389 return &cmd->hw_stats; 390 } else { 391 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va; 392 393 return &cmd->hw_stats; 394 } 395 } 396 397 /* BE2 supports only v0 cmd */ 398 static void *be_erx_stats_from_cmd(struct be_adapter *adapter) 399 { 400 if (BE2_chip(adapter)) { 401 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter); 402 403 return &hw_stats->erx; 404 } else if (BE3_chip(adapter)) { 405 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter); 406 407 return &hw_stats->erx; 408 } else { 409 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter); 410 411 return &hw_stats->erx; 412 } 413 } 414 415 static void populate_be_v0_stats(struct be_adapter *adapter) 416 { 417 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter); 418 struct be_pmem_stats *pmem_sts = &hw_stats->pmem; 419 struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf; 420 struct be_port_rxf_stats_v0 *port_stats = 421 &rxf_stats->port[adapter->port_num]; 422 struct be_drv_stats *drvs = &adapter->drv_stats; 423 424 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats)); 425 drvs->rx_pause_frames = port_stats->rx_pause_frames; 426 drvs->rx_crc_errors = port_stats->rx_crc_errors; 427 drvs->rx_control_frames = port_stats->rx_control_frames; 428 drvs->rx_in_range_errors = port_stats->rx_in_range_errors; 429 drvs->rx_frame_too_long = port_stats->rx_frame_too_long; 430 drvs->rx_dropped_runt = port_stats->rx_dropped_runt; 431 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs; 432 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs; 433 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs; 434 drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow; 435 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length; 436 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small; 437 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short; 438 drvs->rx_out_range_errors = port_stats->rx_out_range_errors; 439 drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow; 440 drvs->rx_dropped_header_too_small = 441 port_stats->rx_dropped_header_too_small; 442 drvs->rx_address_filtered = 443 port_stats->rx_address_filtered + 444 port_stats->rx_vlan_filtered; 445 drvs->rx_alignment_symbol_errors = 446 port_stats->rx_alignment_symbol_errors; 447 448 drvs->tx_pauseframes = port_stats->tx_pauseframes; 449 drvs->tx_controlframes = port_stats->tx_controlframes; 450 451 if (adapter->port_num) 452 drvs->jabber_events = rxf_stats->port1_jabber_events; 453 else 454 drvs->jabber_events = rxf_stats->port0_jabber_events; 455 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf; 456 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr; 457 drvs->forwarded_packets = rxf_stats->forwarded_packets; 458 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu; 459 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr; 460 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags; 461 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops; 462 } 463 464 static void populate_be_v1_stats(struct be_adapter *adapter) 465 { 466 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter); 467 struct be_pmem_stats *pmem_sts = &hw_stats->pmem; 468 struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf; 469 struct be_port_rxf_stats_v1 *port_stats = 470 &rxf_stats->port[adapter->port_num]; 471 struct be_drv_stats *drvs = &adapter->drv_stats; 472 473 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats)); 474 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop; 475 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames; 476 drvs->rx_pause_frames = port_stats->rx_pause_frames; 477 drvs->rx_crc_errors = port_stats->rx_crc_errors; 478 drvs->rx_control_frames = port_stats->rx_control_frames; 479 drvs->rx_in_range_errors = port_stats->rx_in_range_errors; 480 drvs->rx_frame_too_long = port_stats->rx_frame_too_long; 481 drvs->rx_dropped_runt = port_stats->rx_dropped_runt; 482 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs; 483 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs; 484 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs; 485 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length; 486 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small; 487 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short; 488 drvs->rx_out_range_errors = port_stats->rx_out_range_errors; 489 drvs->rx_dropped_header_too_small = 490 port_stats->rx_dropped_header_too_small; 491 drvs->rx_input_fifo_overflow_drop = 492 port_stats->rx_input_fifo_overflow_drop; 493 drvs->rx_address_filtered = port_stats->rx_address_filtered; 494 drvs->rx_alignment_symbol_errors = 495 port_stats->rx_alignment_symbol_errors; 496 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop; 497 drvs->tx_pauseframes = port_stats->tx_pauseframes; 498 drvs->tx_controlframes = port_stats->tx_controlframes; 499 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes; 500 drvs->jabber_events = port_stats->jabber_events; 501 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf; 502 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr; 503 drvs->forwarded_packets = rxf_stats->forwarded_packets; 504 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu; 505 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr; 506 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags; 507 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops; 508 } 509 510 static void populate_be_v2_stats(struct be_adapter *adapter) 511 { 512 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter); 513 struct be_pmem_stats *pmem_sts = &hw_stats->pmem; 514 struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf; 515 struct be_port_rxf_stats_v2 *port_stats = 516 &rxf_stats->port[adapter->port_num]; 517 struct be_drv_stats *drvs = &adapter->drv_stats; 518 519 be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats)); 520 drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop; 521 drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames; 522 drvs->rx_pause_frames = port_stats->rx_pause_frames; 523 drvs->rx_crc_errors = port_stats->rx_crc_errors; 524 drvs->rx_control_frames = port_stats->rx_control_frames; 525 drvs->rx_in_range_errors = port_stats->rx_in_range_errors; 526 drvs->rx_frame_too_long = port_stats->rx_frame_too_long; 527 drvs->rx_dropped_runt = port_stats->rx_dropped_runt; 528 drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs; 529 drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs; 530 drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs; 531 drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length; 532 drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small; 533 drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short; 534 drvs->rx_out_range_errors = port_stats->rx_out_range_errors; 535 drvs->rx_dropped_header_too_small = 536 port_stats->rx_dropped_header_too_small; 537 drvs->rx_input_fifo_overflow_drop = 538 port_stats->rx_input_fifo_overflow_drop; 539 drvs->rx_address_filtered = port_stats->rx_address_filtered; 540 drvs->rx_alignment_symbol_errors = 541 port_stats->rx_alignment_symbol_errors; 542 drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop; 543 drvs->tx_pauseframes = port_stats->tx_pauseframes; 544 drvs->tx_controlframes = port_stats->tx_controlframes; 545 drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes; 546 drvs->jabber_events = port_stats->jabber_events; 547 drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf; 548 drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr; 549 drvs->forwarded_packets = rxf_stats->forwarded_packets; 550 drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu; 551 drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr; 552 drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags; 553 adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops; 554 if (be_roce_supported(adapter)) { 555 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd; 556 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd; 557 drvs->rx_roce_frames = port_stats->roce_frames_received; 558 drvs->roce_drops_crc = port_stats->roce_drops_crc; 559 drvs->roce_drops_payload_len = 560 port_stats->roce_drops_payload_len; 561 } 562 } 563 564 static void populate_lancer_stats(struct be_adapter *adapter) 565 { 566 struct be_drv_stats *drvs = &adapter->drv_stats; 567 struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter); 568 569 be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats)); 570 drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo; 571 drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo; 572 drvs->rx_control_frames = pport_stats->rx_control_frames_lo; 573 drvs->rx_in_range_errors = pport_stats->rx_in_range_errors; 574 drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo; 575 drvs->rx_dropped_runt = pport_stats->rx_dropped_runt; 576 drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors; 577 drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors; 578 drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors; 579 drvs->rx_dropped_tcp_length = 580 pport_stats->rx_dropped_invalid_tcp_length; 581 drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small; 582 drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short; 583 drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors; 584 drvs->rx_dropped_header_too_small = 585 pport_stats->rx_dropped_header_too_small; 586 drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow; 587 drvs->rx_address_filtered = 588 pport_stats->rx_address_filtered + 589 pport_stats->rx_vlan_filtered; 590 drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo; 591 drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow; 592 drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo; 593 drvs->tx_controlframes = pport_stats->tx_control_frames_lo; 594 drvs->jabber_events = pport_stats->rx_jabbers; 595 drvs->forwarded_packets = pport_stats->num_forwards_lo; 596 drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo; 597 drvs->rx_drops_too_many_frags = 598 pport_stats->rx_drops_too_many_frags_lo; 599 } 600 601 static void accumulate_16bit_val(u32 *acc, u16 val) 602 { 603 #define lo(x) (x & 0xFFFF) 604 #define hi(x) (x & 0xFFFF0000) 605 bool wrapped = val < lo(*acc); 606 u32 newacc = hi(*acc) + val; 607 608 if (wrapped) 609 newacc += 65536; 610 WRITE_ONCE(*acc, newacc); 611 } 612 613 static void populate_erx_stats(struct be_adapter *adapter, 614 struct be_rx_obj *rxo, u32 erx_stat) 615 { 616 if (!BEx_chip(adapter)) 617 rx_stats(rxo)->rx_drops_no_frags = erx_stat; 618 else 619 /* below erx HW counter can actually wrap around after 620 * 65535. Driver accumulates a 32-bit value 621 */ 622 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags, 623 (u16)erx_stat); 624 } 625 626 void be_parse_stats(struct be_adapter *adapter) 627 { 628 struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter); 629 struct be_rx_obj *rxo; 630 int i; 631 u32 erx_stat; 632 633 if (lancer_chip(adapter)) { 634 populate_lancer_stats(adapter); 635 } else { 636 if (BE2_chip(adapter)) 637 populate_be_v0_stats(adapter); 638 else if (BE3_chip(adapter)) 639 /* for BE3 */ 640 populate_be_v1_stats(adapter); 641 else 642 populate_be_v2_stats(adapter); 643 644 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */ 645 for_all_rx_queues(adapter, rxo, i) { 646 erx_stat = erx->rx_drops_no_fragments[rxo->q.id]; 647 populate_erx_stats(adapter, rxo, erx_stat); 648 } 649 } 650 } 651 652 static void be_get_stats64(struct net_device *netdev, 653 struct rtnl_link_stats64 *stats) 654 { 655 struct be_adapter *adapter = netdev_priv(netdev); 656 struct be_drv_stats *drvs = &adapter->drv_stats; 657 struct be_rx_obj *rxo; 658 struct be_tx_obj *txo; 659 u64 pkts, bytes; 660 unsigned int start; 661 int i; 662 663 for_all_rx_queues(adapter, rxo, i) { 664 const struct be_rx_stats *rx_stats = rx_stats(rxo); 665 666 do { 667 start = u64_stats_fetch_begin(&rx_stats->sync); 668 pkts = rx_stats(rxo)->rx_pkts; 669 bytes = rx_stats(rxo)->rx_bytes; 670 } while (u64_stats_fetch_retry(&rx_stats->sync, start)); 671 stats->rx_packets += pkts; 672 stats->rx_bytes += bytes; 673 stats->multicast += rx_stats(rxo)->rx_mcast_pkts; 674 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs + 675 rx_stats(rxo)->rx_drops_no_frags; 676 } 677 678 for_all_tx_queues(adapter, txo, i) { 679 const struct be_tx_stats *tx_stats = tx_stats(txo); 680 681 do { 682 start = u64_stats_fetch_begin(&tx_stats->sync); 683 pkts = tx_stats(txo)->tx_pkts; 684 bytes = tx_stats(txo)->tx_bytes; 685 } while (u64_stats_fetch_retry(&tx_stats->sync, start)); 686 stats->tx_packets += pkts; 687 stats->tx_bytes += bytes; 688 } 689 690 /* bad pkts received */ 691 stats->rx_errors = drvs->rx_crc_errors + 692 drvs->rx_alignment_symbol_errors + 693 drvs->rx_in_range_errors + 694 drvs->rx_out_range_errors + 695 drvs->rx_frame_too_long + 696 drvs->rx_dropped_too_small + 697 drvs->rx_dropped_too_short + 698 drvs->rx_dropped_header_too_small + 699 drvs->rx_dropped_tcp_length + 700 drvs->rx_dropped_runt; 701 702 /* detailed rx errors */ 703 stats->rx_length_errors = drvs->rx_in_range_errors + 704 drvs->rx_out_range_errors + 705 drvs->rx_frame_too_long; 706 707 stats->rx_crc_errors = drvs->rx_crc_errors; 708 709 /* frame alignment errors */ 710 stats->rx_frame_errors = drvs->rx_alignment_symbol_errors; 711 712 /* receiver fifo overrun */ 713 /* drops_no_pbuf is no per i/f, it's per BE card */ 714 stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop + 715 drvs->rx_input_fifo_overflow_drop + 716 drvs->rx_drops_no_pbuf; 717 } 718 719 void be_link_status_update(struct be_adapter *adapter, u8 link_status) 720 { 721 struct net_device *netdev = adapter->netdev; 722 723 if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) { 724 netif_carrier_off(netdev); 725 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT; 726 } 727 728 if (link_status) 729 netif_carrier_on(netdev); 730 else 731 netif_carrier_off(netdev); 732 733 netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down"); 734 } 735 736 static int be_gso_hdr_len(struct sk_buff *skb) 737 { 738 if (skb->encapsulation) 739 return skb_inner_tcp_all_headers(skb); 740 741 return skb_tcp_all_headers(skb); 742 } 743 744 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb) 745 { 746 struct be_tx_stats *stats = tx_stats(txo); 747 u32 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1; 748 /* Account for headers which get duplicated in TSO pkt */ 749 u32 dup_hdr_len = tx_pkts > 1 ? be_gso_hdr_len(skb) * (tx_pkts - 1) : 0; 750 751 u64_stats_update_begin(&stats->sync); 752 stats->tx_reqs++; 753 stats->tx_bytes += skb->len + dup_hdr_len; 754 stats->tx_pkts += tx_pkts; 755 if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL) 756 stats->tx_vxlan_offload_pkts += tx_pkts; 757 u64_stats_update_end(&stats->sync); 758 } 759 760 /* Returns number of WRBs needed for the skb */ 761 static u32 skb_wrb_cnt(struct sk_buff *skb) 762 { 763 /* +1 for the header wrb */ 764 return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags; 765 } 766 767 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len) 768 { 769 wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr)); 770 wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr)); 771 wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK); 772 wrb->rsvd0 = 0; 773 } 774 775 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb 776 * to avoid the swap and shift/mask operations in wrb_fill(). 777 */ 778 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb) 779 { 780 wrb->frag_pa_hi = 0; 781 wrb->frag_pa_lo = 0; 782 wrb->frag_len = 0; 783 wrb->rsvd0 = 0; 784 } 785 786 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter, 787 struct sk_buff *skb) 788 { 789 u8 vlan_prio; 790 u16 vlan_tag; 791 792 vlan_tag = skb_vlan_tag_get(skb); 793 vlan_prio = skb_vlan_tag_get_prio(skb); 794 /* If vlan priority provided by OS is NOT in available bmap */ 795 if (!(adapter->vlan_prio_bmap & (1 << vlan_prio))) 796 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) | 797 adapter->recommended_prio_bits; 798 799 return vlan_tag; 800 } 801 802 /* Used only for IP tunnel packets */ 803 static u16 skb_inner_ip_proto(struct sk_buff *skb) 804 { 805 return (inner_ip_hdr(skb)->version == 4) ? 806 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr; 807 } 808 809 static u16 skb_ip_proto(struct sk_buff *skb) 810 { 811 return (ip_hdr(skb)->version == 4) ? 812 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr; 813 } 814 815 static inline bool be_is_txq_full(struct be_tx_obj *txo) 816 { 817 return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len; 818 } 819 820 static inline bool be_can_txq_wake(struct be_tx_obj *txo) 821 { 822 return atomic_read(&txo->q.used) < txo->q.len / 2; 823 } 824 825 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo) 826 { 827 return atomic_read(&txo->q.used) > txo->pend_wrb_cnt; 828 } 829 830 static void be_get_wrb_params_from_skb(struct be_adapter *adapter, 831 struct sk_buff *skb, 832 struct be_wrb_params *wrb_params) 833 { 834 u16 proto; 835 836 if (skb_is_gso(skb)) { 837 BE_WRB_F_SET(wrb_params->features, LSO, 1); 838 wrb_params->lso_mss = skb_shinfo(skb)->gso_size; 839 if (skb_is_gso_v6(skb) && !lancer_chip(adapter)) 840 BE_WRB_F_SET(wrb_params->features, LSO6, 1); 841 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 842 if (skb->encapsulation) { 843 BE_WRB_F_SET(wrb_params->features, IPCS, 1); 844 proto = skb_inner_ip_proto(skb); 845 } else { 846 proto = skb_ip_proto(skb); 847 } 848 if (proto == IPPROTO_TCP) 849 BE_WRB_F_SET(wrb_params->features, TCPCS, 1); 850 else if (proto == IPPROTO_UDP) 851 BE_WRB_F_SET(wrb_params->features, UDPCS, 1); 852 } 853 854 if (skb_vlan_tag_present(skb)) { 855 BE_WRB_F_SET(wrb_params->features, VLAN, 1); 856 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb); 857 } 858 859 BE_WRB_F_SET(wrb_params->features, CRC, 1); 860 } 861 862 static void wrb_fill_hdr(struct be_adapter *adapter, 863 struct be_eth_hdr_wrb *hdr, 864 struct be_wrb_params *wrb_params, 865 struct sk_buff *skb) 866 { 867 memset(hdr, 0, sizeof(*hdr)); 868 869 SET_TX_WRB_HDR_BITS(crc, hdr, 870 BE_WRB_F_GET(wrb_params->features, CRC)); 871 SET_TX_WRB_HDR_BITS(ipcs, hdr, 872 BE_WRB_F_GET(wrb_params->features, IPCS)); 873 SET_TX_WRB_HDR_BITS(tcpcs, hdr, 874 BE_WRB_F_GET(wrb_params->features, TCPCS)); 875 SET_TX_WRB_HDR_BITS(udpcs, hdr, 876 BE_WRB_F_GET(wrb_params->features, UDPCS)); 877 878 SET_TX_WRB_HDR_BITS(lso, hdr, 879 BE_WRB_F_GET(wrb_params->features, LSO)); 880 SET_TX_WRB_HDR_BITS(lso6, hdr, 881 BE_WRB_F_GET(wrb_params->features, LSO6)); 882 SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss); 883 884 /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this 885 * hack is not needed, the evt bit is set while ringing DB. 886 */ 887 SET_TX_WRB_HDR_BITS(event, hdr, 888 BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW)); 889 SET_TX_WRB_HDR_BITS(vlan, hdr, 890 BE_WRB_F_GET(wrb_params->features, VLAN)); 891 SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag); 892 893 SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb)); 894 SET_TX_WRB_HDR_BITS(len, hdr, skb->len); 895 SET_TX_WRB_HDR_BITS(mgmt, hdr, 896 BE_WRB_F_GET(wrb_params->features, OS2BMC)); 897 } 898 899 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb, 900 bool unmap_single) 901 { 902 dma_addr_t dma; 903 u32 frag_len = le32_to_cpu(wrb->frag_len); 904 905 906 dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 | 907 (u64)le32_to_cpu(wrb->frag_pa_lo); 908 if (frag_len) { 909 if (unmap_single) 910 dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE); 911 else 912 dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE); 913 } 914 } 915 916 /* Grab a WRB header for xmit */ 917 static u32 be_tx_get_wrb_hdr(struct be_tx_obj *txo) 918 { 919 u32 head = txo->q.head; 920 921 queue_head_inc(&txo->q); 922 return head; 923 } 924 925 /* Set up the WRB header for xmit */ 926 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter, 927 struct be_tx_obj *txo, 928 struct be_wrb_params *wrb_params, 929 struct sk_buff *skb, u16 head) 930 { 931 u32 num_frags = skb_wrb_cnt(skb); 932 struct be_queue_info *txq = &txo->q; 933 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head); 934 935 wrb_fill_hdr(adapter, hdr, wrb_params, skb); 936 be_dws_cpu_to_le(hdr, sizeof(*hdr)); 937 938 BUG_ON(txo->sent_skb_list[head]); 939 txo->sent_skb_list[head] = skb; 940 txo->last_req_hdr = head; 941 atomic_add(num_frags, &txq->used); 942 txo->last_req_wrb_cnt = num_frags; 943 txo->pend_wrb_cnt += num_frags; 944 } 945 946 /* Setup a WRB fragment (buffer descriptor) for xmit */ 947 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr, 948 int len) 949 { 950 struct be_eth_wrb *wrb; 951 struct be_queue_info *txq = &txo->q; 952 953 wrb = queue_head_node(txq); 954 wrb_fill(wrb, busaddr, len); 955 queue_head_inc(txq); 956 } 957 958 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine 959 * was invoked. The producer index is restored to the previous packet and the 960 * WRBs of the current packet are unmapped. Invoked to handle tx setup errors. 961 */ 962 static void be_xmit_restore(struct be_adapter *adapter, 963 struct be_tx_obj *txo, u32 head, bool map_single, 964 u32 copied) 965 { 966 struct device *dev; 967 struct be_eth_wrb *wrb; 968 struct be_queue_info *txq = &txo->q; 969 970 dev = &adapter->pdev->dev; 971 txq->head = head; 972 973 /* skip the first wrb (hdr); it's not mapped */ 974 queue_head_inc(txq); 975 while (copied) { 976 wrb = queue_head_node(txq); 977 unmap_tx_frag(dev, wrb, map_single); 978 map_single = false; 979 copied -= le32_to_cpu(wrb->frag_len); 980 queue_head_inc(txq); 981 } 982 983 txq->head = head; 984 } 985 986 /* Enqueue the given packet for transmit. This routine allocates WRBs for the 987 * packet, dma maps the packet buffers and sets up the WRBs. Returns the number 988 * of WRBs used up by the packet. 989 */ 990 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo, 991 struct sk_buff *skb, 992 struct be_wrb_params *wrb_params) 993 { 994 u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb); 995 struct device *dev = &adapter->pdev->dev; 996 bool map_single = false; 997 u32 head; 998 dma_addr_t busaddr; 999 int len; 1000 1001 head = be_tx_get_wrb_hdr(txo); 1002 1003 if (skb->len > skb->data_len) { 1004 len = skb_headlen(skb); 1005 1006 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE); 1007 if (dma_mapping_error(dev, busaddr)) 1008 goto dma_err; 1009 map_single = true; 1010 be_tx_setup_wrb_frag(txo, busaddr, len); 1011 copied += len; 1012 } 1013 1014 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1015 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1016 len = skb_frag_size(frag); 1017 1018 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE); 1019 if (dma_mapping_error(dev, busaddr)) 1020 goto dma_err; 1021 be_tx_setup_wrb_frag(txo, busaddr, len); 1022 copied += len; 1023 } 1024 1025 be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head); 1026 1027 be_tx_stats_update(txo, skb); 1028 return wrb_cnt; 1029 1030 dma_err: 1031 adapter->drv_stats.dma_map_errors++; 1032 be_xmit_restore(adapter, txo, head, map_single, copied); 1033 return 0; 1034 } 1035 1036 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter) 1037 { 1038 return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD; 1039 } 1040 1041 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter, 1042 struct sk_buff *skb, 1043 struct be_wrb_params 1044 *wrb_params) 1045 { 1046 bool insert_vlan = false; 1047 u16 vlan_tag = 0; 1048 1049 skb = skb_share_check(skb, GFP_ATOMIC); 1050 if (unlikely(!skb)) 1051 return skb; 1052 1053 if (skb_vlan_tag_present(skb)) { 1054 vlan_tag = be_get_tx_vlan_tag(adapter, skb); 1055 insert_vlan = true; 1056 } 1057 1058 if (qnq_async_evt_rcvd(adapter) && adapter->pvid) { 1059 if (!insert_vlan) { 1060 vlan_tag = adapter->pvid; 1061 insert_vlan = true; 1062 } 1063 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to 1064 * skip VLAN insertion 1065 */ 1066 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1); 1067 } 1068 1069 if (insert_vlan) { 1070 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q), 1071 vlan_tag); 1072 if (unlikely(!skb)) 1073 return skb; 1074 __vlan_hwaccel_clear_tag(skb); 1075 } 1076 1077 /* Insert the outer VLAN, if any */ 1078 if (adapter->qnq_vid) { 1079 vlan_tag = adapter->qnq_vid; 1080 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q), 1081 vlan_tag); 1082 if (unlikely(!skb)) 1083 return skb; 1084 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1); 1085 } 1086 1087 return skb; 1088 } 1089 1090 static bool be_ipv6_exthdr_check(struct sk_buff *skb) 1091 { 1092 struct ethhdr *eh = (struct ethhdr *)skb->data; 1093 u16 offset = ETH_HLEN; 1094 1095 if (eh->h_proto == htons(ETH_P_IPV6)) { 1096 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset); 1097 1098 offset += sizeof(struct ipv6hdr); 1099 if (ip6h->nexthdr != NEXTHDR_TCP && 1100 ip6h->nexthdr != NEXTHDR_UDP) { 1101 struct ipv6_opt_hdr *ehdr = 1102 (struct ipv6_opt_hdr *)(skb->data + offset); 1103 1104 /* offending pkt: 2nd byte following IPv6 hdr is 0xff */ 1105 if (ehdr->hdrlen == 0xff) 1106 return true; 1107 } 1108 } 1109 return false; 1110 } 1111 1112 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb) 1113 { 1114 return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid; 1115 } 1116 1117 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb) 1118 { 1119 return BE3_chip(adapter) && be_ipv6_exthdr_check(skb); 1120 } 1121 1122 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter, 1123 struct sk_buff *skb, 1124 struct be_wrb_params 1125 *wrb_params) 1126 { 1127 struct vlan_ethhdr *veh = skb_vlan_eth_hdr(skb); 1128 unsigned int eth_hdr_len; 1129 struct iphdr *ip; 1130 1131 /* For padded packets, BE HW modifies tot_len field in IP header 1132 * incorrectly when VLAN tag is inserted by HW. 1133 * For padded packets, Lancer computes incorrect checksum. 1134 */ 1135 eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ? 1136 VLAN_ETH_HLEN : ETH_HLEN; 1137 if (skb->len <= 60 && 1138 (lancer_chip(adapter) || BE3_chip(adapter) || 1139 skb_vlan_tag_present(skb)) && is_ipv4_pkt(skb)) { 1140 ip = (struct iphdr *)ip_hdr(skb); 1141 if (unlikely(pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len)))) 1142 goto tx_drop; 1143 } 1144 1145 /* If vlan tag is already inlined in the packet, skip HW VLAN 1146 * tagging in pvid-tagging mode 1147 */ 1148 if (be_pvid_tagging_enabled(adapter) && 1149 veh->h_vlan_proto == htons(ETH_P_8021Q)) 1150 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1); 1151 1152 /* HW has a bug wherein it will calculate CSUM for VLAN 1153 * pkts even though it is disabled. 1154 * Manually insert VLAN in pkt. 1155 */ 1156 if (skb->ip_summed != CHECKSUM_PARTIAL && 1157 skb_vlan_tag_present(skb)) { 1158 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params); 1159 if (unlikely(!skb)) 1160 goto err; 1161 } 1162 1163 /* HW may lockup when VLAN HW tagging is requested on 1164 * certain ipv6 packets. Drop such pkts if the HW workaround to 1165 * skip HW tagging is not enabled by FW. 1166 */ 1167 if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) && 1168 (adapter->pvid || adapter->qnq_vid) && 1169 !qnq_async_evt_rcvd(adapter))) 1170 goto tx_drop; 1171 1172 /* Manual VLAN tag insertion to prevent: 1173 * ASIC lockup when the ASIC inserts VLAN tag into 1174 * certain ipv6 packets. Insert VLAN tags in driver, 1175 * and set event, completion, vlan bits accordingly 1176 * in the Tx WRB. 1177 */ 1178 if (be_ipv6_tx_stall_chk(adapter, skb) && 1179 be_vlan_tag_tx_chk(adapter, skb)) { 1180 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params); 1181 if (unlikely(!skb)) 1182 goto err; 1183 } 1184 1185 return skb; 1186 tx_drop: 1187 dev_kfree_skb_any(skb); 1188 err: 1189 return NULL; 1190 } 1191 1192 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter, 1193 struct sk_buff *skb, 1194 struct be_wrb_params *wrb_params) 1195 { 1196 int err; 1197 1198 /* Lancer, SH and BE3 in SRIOV mode have a bug wherein 1199 * packets that are 32b or less may cause a transmit stall 1200 * on that port. The workaround is to pad such packets 1201 * (len <= 32 bytes) to a minimum length of 36b. 1202 */ 1203 if (skb->len <= 32) { 1204 if (skb_put_padto(skb, 36)) 1205 return NULL; 1206 } 1207 1208 if (BEx_chip(adapter) || lancer_chip(adapter)) { 1209 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params); 1210 if (!skb) 1211 return NULL; 1212 } 1213 1214 /* The stack can send us skbs with length greater than 1215 * what the HW can handle. Trim the extra bytes. 1216 */ 1217 WARN_ON_ONCE(skb->len > BE_MAX_GSO_SIZE); 1218 err = pskb_trim(skb, BE_MAX_GSO_SIZE); 1219 WARN_ON(err); 1220 1221 return skb; 1222 } 1223 1224 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo) 1225 { 1226 struct be_queue_info *txq = &txo->q; 1227 struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr); 1228 1229 /* Mark the last request eventable if it hasn't been marked already */ 1230 if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT))) 1231 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL); 1232 1233 /* compose a dummy wrb if there are odd set of wrbs to notify */ 1234 if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) { 1235 wrb_fill_dummy(queue_head_node(txq)); 1236 queue_head_inc(txq); 1237 atomic_inc(&txq->used); 1238 txo->pend_wrb_cnt++; 1239 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK << 1240 TX_HDR_WRB_NUM_SHIFT); 1241 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) << 1242 TX_HDR_WRB_NUM_SHIFT); 1243 } 1244 be_txq_notify(adapter, txo, txo->pend_wrb_cnt); 1245 txo->pend_wrb_cnt = 0; 1246 } 1247 1248 /* OS2BMC related */ 1249 1250 #define DHCP_CLIENT_PORT 68 1251 #define DHCP_SERVER_PORT 67 1252 #define NET_BIOS_PORT1 137 1253 #define NET_BIOS_PORT2 138 1254 #define DHCPV6_RAS_PORT 547 1255 1256 #define is_mc_allowed_on_bmc(adapter, eh) \ 1257 (!is_multicast_filt_enabled(adapter) && \ 1258 is_multicast_ether_addr(eh->h_dest) && \ 1259 !is_broadcast_ether_addr(eh->h_dest)) 1260 1261 #define is_bc_allowed_on_bmc(adapter, eh) \ 1262 (!is_broadcast_filt_enabled(adapter) && \ 1263 is_broadcast_ether_addr(eh->h_dest)) 1264 1265 #define is_arp_allowed_on_bmc(adapter, skb) \ 1266 (is_arp(skb) && is_arp_filt_enabled(adapter)) 1267 1268 #define is_arp(skb) (skb->protocol == htons(ETH_P_ARP)) 1269 1270 #define is_arp_filt_enabled(adapter) \ 1271 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP)) 1272 1273 #define is_dhcp_client_filt_enabled(adapter) \ 1274 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT) 1275 1276 #define is_dhcp_srvr_filt_enabled(adapter) \ 1277 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER) 1278 1279 #define is_nbios_filt_enabled(adapter) \ 1280 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS) 1281 1282 #define is_ipv6_na_filt_enabled(adapter) \ 1283 (adapter->bmc_filt_mask & \ 1284 BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER) 1285 1286 #define is_ipv6_ra_filt_enabled(adapter) \ 1287 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA) 1288 1289 #define is_ipv6_ras_filt_enabled(adapter) \ 1290 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS) 1291 1292 #define is_broadcast_filt_enabled(adapter) \ 1293 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST) 1294 1295 #define is_multicast_filt_enabled(adapter) \ 1296 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST) 1297 1298 static bool be_send_pkt_to_bmc(struct be_adapter *adapter, 1299 struct sk_buff **skb, 1300 struct be_wrb_params *wrb_params) 1301 { 1302 struct ethhdr *eh = (struct ethhdr *)(*skb)->data; 1303 bool os2bmc = false; 1304 1305 if (!be_is_os2bmc_enabled(adapter)) 1306 goto done; 1307 1308 if (!is_multicast_ether_addr(eh->h_dest)) 1309 goto done; 1310 1311 if (is_mc_allowed_on_bmc(adapter, eh) || 1312 is_bc_allowed_on_bmc(adapter, eh) || 1313 is_arp_allowed_on_bmc(adapter, (*skb))) { 1314 os2bmc = true; 1315 goto done; 1316 } 1317 1318 if ((*skb)->protocol == htons(ETH_P_IPV6)) { 1319 struct ipv6hdr *hdr = ipv6_hdr((*skb)); 1320 u8 nexthdr = hdr->nexthdr; 1321 1322 if (nexthdr == IPPROTO_ICMPV6) { 1323 struct icmp6hdr *icmp6 = icmp6_hdr((*skb)); 1324 1325 switch (icmp6->icmp6_type) { 1326 case NDISC_ROUTER_ADVERTISEMENT: 1327 os2bmc = is_ipv6_ra_filt_enabled(adapter); 1328 goto done; 1329 case NDISC_NEIGHBOUR_ADVERTISEMENT: 1330 os2bmc = is_ipv6_na_filt_enabled(adapter); 1331 goto done; 1332 default: 1333 break; 1334 } 1335 } 1336 } 1337 1338 if (is_udp_pkt((*skb))) { 1339 struct udphdr *udp = udp_hdr((*skb)); 1340 1341 switch (ntohs(udp->dest)) { 1342 case DHCP_CLIENT_PORT: 1343 os2bmc = is_dhcp_client_filt_enabled(adapter); 1344 goto done; 1345 case DHCP_SERVER_PORT: 1346 os2bmc = is_dhcp_srvr_filt_enabled(adapter); 1347 goto done; 1348 case NET_BIOS_PORT1: 1349 case NET_BIOS_PORT2: 1350 os2bmc = is_nbios_filt_enabled(adapter); 1351 goto done; 1352 case DHCPV6_RAS_PORT: 1353 os2bmc = is_ipv6_ras_filt_enabled(adapter); 1354 goto done; 1355 default: 1356 break; 1357 } 1358 } 1359 done: 1360 /* For packets over a vlan, which are destined 1361 * to BMC, asic expects the vlan to be inline in the packet. 1362 */ 1363 if (os2bmc) 1364 *skb = be_insert_vlan_in_pkt(adapter, *skb, wrb_params); 1365 1366 return os2bmc; 1367 } 1368 1369 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev) 1370 { 1371 struct be_adapter *adapter = netdev_priv(netdev); 1372 u16 q_idx = skb_get_queue_mapping(skb); 1373 struct be_tx_obj *txo = &adapter->tx_obj[q_idx]; 1374 struct be_wrb_params wrb_params = { 0 }; 1375 bool flush = !netdev_xmit_more(); 1376 u16 wrb_cnt; 1377 1378 skb = be_xmit_workarounds(adapter, skb, &wrb_params); 1379 if (unlikely(!skb)) 1380 goto drop; 1381 1382 be_get_wrb_params_from_skb(adapter, skb, &wrb_params); 1383 1384 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params); 1385 if (unlikely(!wrb_cnt)) 1386 goto drop_skb; 1387 1388 /* if os2bmc is enabled and if the pkt is destined to bmc, 1389 * enqueue the pkt a 2nd time with mgmt bit set. 1390 */ 1391 if (be_send_pkt_to_bmc(adapter, &skb, &wrb_params)) { 1392 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1); 1393 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params); 1394 if (unlikely(!wrb_cnt)) 1395 goto drop_skb; 1396 else 1397 skb_get(skb); 1398 } 1399 1400 if (be_is_txq_full(txo)) { 1401 netif_stop_subqueue(netdev, q_idx); 1402 tx_stats(txo)->tx_stops++; 1403 } 1404 1405 if (flush || __netif_subqueue_stopped(netdev, q_idx)) 1406 be_xmit_flush(adapter, txo); 1407 1408 return NETDEV_TX_OK; 1409 drop_skb: 1410 dev_kfree_skb_any(skb); 1411 drop: 1412 tx_stats(txo)->tx_drv_drops++; 1413 /* Flush the already enqueued tx requests */ 1414 if (flush && txo->pend_wrb_cnt) 1415 be_xmit_flush(adapter, txo); 1416 1417 return NETDEV_TX_OK; 1418 } 1419 1420 static void be_tx_timeout(struct net_device *netdev, unsigned int txqueue) 1421 { 1422 struct be_adapter *adapter = netdev_priv(netdev); 1423 struct device *dev = &adapter->pdev->dev; 1424 struct be_tx_obj *txo; 1425 struct sk_buff *skb; 1426 struct tcphdr *tcphdr; 1427 struct udphdr *udphdr; 1428 u32 *entry; 1429 int status; 1430 int i, j; 1431 1432 for_all_tx_queues(adapter, txo, i) { 1433 dev_info(dev, "TXQ Dump: %d H: %d T: %d used: %d, qid: 0x%x\n", 1434 i, txo->q.head, txo->q.tail, 1435 atomic_read(&txo->q.used), txo->q.id); 1436 1437 entry = txo->q.dma_mem.va; 1438 for (j = 0; j < TX_Q_LEN * 4; j += 4) { 1439 if (entry[j] != 0 || entry[j + 1] != 0 || 1440 entry[j + 2] != 0 || entry[j + 3] != 0) { 1441 dev_info(dev, "Entry %d 0x%x 0x%x 0x%x 0x%x\n", 1442 j, entry[j], entry[j + 1], 1443 entry[j + 2], entry[j + 3]); 1444 } 1445 } 1446 1447 entry = txo->cq.dma_mem.va; 1448 dev_info(dev, "TXCQ Dump: %d H: %d T: %d used: %d\n", 1449 i, txo->cq.head, txo->cq.tail, 1450 atomic_read(&txo->cq.used)); 1451 for (j = 0; j < TX_CQ_LEN * 4; j += 4) { 1452 if (entry[j] != 0 || entry[j + 1] != 0 || 1453 entry[j + 2] != 0 || entry[j + 3] != 0) { 1454 dev_info(dev, "Entry %d 0x%x 0x%x 0x%x 0x%x\n", 1455 j, entry[j], entry[j + 1], 1456 entry[j + 2], entry[j + 3]); 1457 } 1458 } 1459 1460 for (j = 0; j < TX_Q_LEN; j++) { 1461 if (txo->sent_skb_list[j]) { 1462 skb = txo->sent_skb_list[j]; 1463 if (ip_hdr(skb)->protocol == IPPROTO_TCP) { 1464 tcphdr = tcp_hdr(skb); 1465 dev_info(dev, "TCP source port %d\n", 1466 ntohs(tcphdr->source)); 1467 dev_info(dev, "TCP dest port %d\n", 1468 ntohs(tcphdr->dest)); 1469 dev_info(dev, "TCP sequence num %u\n", 1470 ntohl(tcphdr->seq)); 1471 dev_info(dev, "TCP ack_seq %u\n", 1472 ntohl(tcphdr->ack_seq)); 1473 } else if (ip_hdr(skb)->protocol == 1474 IPPROTO_UDP) { 1475 udphdr = udp_hdr(skb); 1476 dev_info(dev, "UDP source port %d\n", 1477 ntohs(udphdr->source)); 1478 dev_info(dev, "UDP dest port %d\n", 1479 ntohs(udphdr->dest)); 1480 } 1481 dev_info(dev, "skb[%d] %p len %d proto 0x%x\n", 1482 j, skb, skb->len, skb->protocol); 1483 } 1484 } 1485 } 1486 1487 if (lancer_chip(adapter)) { 1488 dev_info(dev, "Initiating reset due to tx timeout\n"); 1489 dev_info(dev, "Resetting adapter\n"); 1490 status = lancer_physdev_ctrl(adapter, 1491 PHYSDEV_CONTROL_FW_RESET_MASK); 1492 if (status) 1493 dev_err(dev, "Reset failed .. Reboot server\n"); 1494 } 1495 } 1496 1497 static inline bool be_in_all_promisc(struct be_adapter *adapter) 1498 { 1499 return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) == 1500 BE_IF_FLAGS_ALL_PROMISCUOUS; 1501 } 1502 1503 static int be_set_vlan_promisc(struct be_adapter *adapter) 1504 { 1505 struct device *dev = &adapter->pdev->dev; 1506 int status; 1507 1508 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) 1509 return 0; 1510 1511 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON); 1512 if (!status) { 1513 dev_info(dev, "Enabled VLAN promiscuous mode\n"); 1514 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS; 1515 } else { 1516 dev_err(dev, "Failed to enable VLAN promiscuous mode\n"); 1517 } 1518 return status; 1519 } 1520 1521 static int be_clear_vlan_promisc(struct be_adapter *adapter) 1522 { 1523 struct device *dev = &adapter->pdev->dev; 1524 int status; 1525 1526 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF); 1527 if (!status) { 1528 dev_info(dev, "Disabling VLAN promiscuous mode\n"); 1529 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS; 1530 } 1531 return status; 1532 } 1533 1534 /* 1535 * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE. 1536 * If the user configures more, place BE in vlan promiscuous mode. 1537 */ 1538 static int be_vid_config(struct be_adapter *adapter) 1539 { 1540 struct device *dev = &adapter->pdev->dev; 1541 u16 vids[BE_NUM_VLANS_SUPPORTED]; 1542 u16 num = 0, i = 0; 1543 int status = 0; 1544 1545 /* No need to change the VLAN state if the I/F is in promiscuous */ 1546 if (adapter->netdev->flags & IFF_PROMISC) 1547 return 0; 1548 1549 if (adapter->vlans_added > be_max_vlans(adapter)) 1550 return be_set_vlan_promisc(adapter); 1551 1552 if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) { 1553 status = be_clear_vlan_promisc(adapter); 1554 if (status) 1555 return status; 1556 } 1557 /* Construct VLAN Table to give to HW */ 1558 for_each_set_bit(i, adapter->vids, VLAN_N_VID) 1559 vids[num++] = cpu_to_le16(i); 1560 1561 status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0); 1562 if (status) { 1563 dev_err(dev, "Setting HW VLAN filtering failed\n"); 1564 /* Set to VLAN promisc mode as setting VLAN filter failed */ 1565 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS || 1566 addl_status(status) == 1567 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES) 1568 return be_set_vlan_promisc(adapter); 1569 } 1570 return status; 1571 } 1572 1573 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid) 1574 { 1575 struct be_adapter *adapter = netdev_priv(netdev); 1576 int status = 0; 1577 1578 mutex_lock(&adapter->rx_filter_lock); 1579 1580 /* Packets with VID 0 are always received by Lancer by default */ 1581 if (lancer_chip(adapter) && vid == 0) 1582 goto done; 1583 1584 if (test_bit(vid, adapter->vids)) 1585 goto done; 1586 1587 set_bit(vid, adapter->vids); 1588 adapter->vlans_added++; 1589 1590 status = be_vid_config(adapter); 1591 done: 1592 mutex_unlock(&adapter->rx_filter_lock); 1593 return status; 1594 } 1595 1596 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid) 1597 { 1598 struct be_adapter *adapter = netdev_priv(netdev); 1599 int status = 0; 1600 1601 mutex_lock(&adapter->rx_filter_lock); 1602 1603 /* Packets with VID 0 are always received by Lancer by default */ 1604 if (lancer_chip(adapter) && vid == 0) 1605 goto done; 1606 1607 if (!test_bit(vid, adapter->vids)) 1608 goto done; 1609 1610 clear_bit(vid, adapter->vids); 1611 adapter->vlans_added--; 1612 1613 status = be_vid_config(adapter); 1614 done: 1615 mutex_unlock(&adapter->rx_filter_lock); 1616 return status; 1617 } 1618 1619 static void be_set_all_promisc(struct be_adapter *adapter) 1620 { 1621 be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON); 1622 adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS; 1623 } 1624 1625 static void be_set_mc_promisc(struct be_adapter *adapter) 1626 { 1627 int status; 1628 1629 if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) 1630 return; 1631 1632 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON); 1633 if (!status) 1634 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS; 1635 } 1636 1637 static void be_set_uc_promisc(struct be_adapter *adapter) 1638 { 1639 int status; 1640 1641 if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) 1642 return; 1643 1644 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, ON); 1645 if (!status) 1646 adapter->if_flags |= BE_IF_FLAGS_PROMISCUOUS; 1647 } 1648 1649 static void be_clear_uc_promisc(struct be_adapter *adapter) 1650 { 1651 int status; 1652 1653 if (!(adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS)) 1654 return; 1655 1656 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_PROMISCUOUS, OFF); 1657 if (!status) 1658 adapter->if_flags &= ~BE_IF_FLAGS_PROMISCUOUS; 1659 } 1660 1661 /* The below 2 functions are the callback args for __dev_mc_sync/dev_uc_sync(). 1662 * We use a single callback function for both sync and unsync. We really don't 1663 * add/remove addresses through this callback. But, we use it to detect changes 1664 * to the uc/mc lists. The entire uc/mc list is programmed in be_set_rx_mode(). 1665 */ 1666 static int be_uc_list_update(struct net_device *netdev, 1667 const unsigned char *addr) 1668 { 1669 struct be_adapter *adapter = netdev_priv(netdev); 1670 1671 adapter->update_uc_list = true; 1672 return 0; 1673 } 1674 1675 static int be_mc_list_update(struct net_device *netdev, 1676 const unsigned char *addr) 1677 { 1678 struct be_adapter *adapter = netdev_priv(netdev); 1679 1680 adapter->update_mc_list = true; 1681 return 0; 1682 } 1683 1684 static void be_set_mc_list(struct be_adapter *adapter) 1685 { 1686 struct net_device *netdev = adapter->netdev; 1687 struct netdev_hw_addr *ha; 1688 bool mc_promisc = false; 1689 int status; 1690 1691 netif_addr_lock_bh(netdev); 1692 __dev_mc_sync(netdev, be_mc_list_update, be_mc_list_update); 1693 1694 if (netdev->flags & IFF_PROMISC) { 1695 adapter->update_mc_list = false; 1696 } else if (netdev->flags & IFF_ALLMULTI || 1697 netdev_mc_count(netdev) > be_max_mc(adapter)) { 1698 /* Enable multicast promisc if num configured exceeds 1699 * what we support 1700 */ 1701 mc_promisc = true; 1702 adapter->update_mc_list = false; 1703 } else if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS) { 1704 /* Update mc-list unconditionally if the iface was previously 1705 * in mc-promisc mode and now is out of that mode. 1706 */ 1707 adapter->update_mc_list = true; 1708 } 1709 1710 if (adapter->update_mc_list) { 1711 int i = 0; 1712 1713 /* cache the mc-list in adapter */ 1714 netdev_for_each_mc_addr(ha, netdev) { 1715 ether_addr_copy(adapter->mc_list[i].mac, ha->addr); 1716 i++; 1717 } 1718 adapter->mc_count = netdev_mc_count(netdev); 1719 } 1720 netif_addr_unlock_bh(netdev); 1721 1722 if (mc_promisc) { 1723 be_set_mc_promisc(adapter); 1724 } else if (adapter->update_mc_list) { 1725 status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON); 1726 if (!status) 1727 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS; 1728 else 1729 be_set_mc_promisc(adapter); 1730 1731 adapter->update_mc_list = false; 1732 } 1733 } 1734 1735 static void be_clear_mc_list(struct be_adapter *adapter) 1736 { 1737 struct net_device *netdev = adapter->netdev; 1738 1739 __dev_mc_unsync(netdev, NULL); 1740 be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, OFF); 1741 adapter->mc_count = 0; 1742 } 1743 1744 static int be_uc_mac_add(struct be_adapter *adapter, int uc_idx) 1745 { 1746 if (ether_addr_equal(adapter->uc_list[uc_idx].mac, adapter->dev_mac)) { 1747 adapter->pmac_id[uc_idx + 1] = adapter->pmac_id[0]; 1748 return 0; 1749 } 1750 1751 return be_cmd_pmac_add(adapter, adapter->uc_list[uc_idx].mac, 1752 adapter->if_handle, 1753 &adapter->pmac_id[uc_idx + 1], 0); 1754 } 1755 1756 static void be_uc_mac_del(struct be_adapter *adapter, int pmac_id) 1757 { 1758 if (pmac_id == adapter->pmac_id[0]) 1759 return; 1760 1761 be_cmd_pmac_del(adapter, adapter->if_handle, pmac_id, 0); 1762 } 1763 1764 static void be_set_uc_list(struct be_adapter *adapter) 1765 { 1766 struct net_device *netdev = adapter->netdev; 1767 struct netdev_hw_addr *ha; 1768 bool uc_promisc = false; 1769 int curr_uc_macs = 0, i; 1770 1771 netif_addr_lock_bh(netdev); 1772 __dev_uc_sync(netdev, be_uc_list_update, be_uc_list_update); 1773 1774 if (netdev->flags & IFF_PROMISC) { 1775 adapter->update_uc_list = false; 1776 } else if (netdev_uc_count(netdev) > (be_max_uc(adapter) - 1)) { 1777 uc_promisc = true; 1778 adapter->update_uc_list = false; 1779 } else if (adapter->if_flags & BE_IF_FLAGS_PROMISCUOUS) { 1780 /* Update uc-list unconditionally if the iface was previously 1781 * in uc-promisc mode and now is out of that mode. 1782 */ 1783 adapter->update_uc_list = true; 1784 } 1785 1786 if (adapter->update_uc_list) { 1787 /* cache the uc-list in adapter array */ 1788 i = 0; 1789 netdev_for_each_uc_addr(ha, netdev) { 1790 ether_addr_copy(adapter->uc_list[i].mac, ha->addr); 1791 i++; 1792 } 1793 curr_uc_macs = netdev_uc_count(netdev); 1794 } 1795 netif_addr_unlock_bh(netdev); 1796 1797 if (uc_promisc) { 1798 be_set_uc_promisc(adapter); 1799 } else if (adapter->update_uc_list) { 1800 be_clear_uc_promisc(adapter); 1801 1802 for (i = 0; i < adapter->uc_macs; i++) 1803 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]); 1804 1805 for (i = 0; i < curr_uc_macs; i++) 1806 be_uc_mac_add(adapter, i); 1807 adapter->uc_macs = curr_uc_macs; 1808 adapter->update_uc_list = false; 1809 } 1810 } 1811 1812 static void be_clear_uc_list(struct be_adapter *adapter) 1813 { 1814 struct net_device *netdev = adapter->netdev; 1815 int i; 1816 1817 __dev_uc_unsync(netdev, NULL); 1818 for (i = 0; i < adapter->uc_macs; i++) 1819 be_uc_mac_del(adapter, adapter->pmac_id[i + 1]); 1820 1821 adapter->uc_macs = 0; 1822 } 1823 1824 static void __be_set_rx_mode(struct be_adapter *adapter) 1825 { 1826 struct net_device *netdev = adapter->netdev; 1827 1828 mutex_lock(&adapter->rx_filter_lock); 1829 1830 if (netdev->flags & IFF_PROMISC) { 1831 if (!be_in_all_promisc(adapter)) 1832 be_set_all_promisc(adapter); 1833 } else if (be_in_all_promisc(adapter)) { 1834 /* We need to re-program the vlan-list or clear 1835 * vlan-promisc mode (if needed) when the interface 1836 * comes out of promisc mode. 1837 */ 1838 be_vid_config(adapter); 1839 } 1840 1841 be_set_uc_list(adapter); 1842 be_set_mc_list(adapter); 1843 1844 mutex_unlock(&adapter->rx_filter_lock); 1845 } 1846 1847 static void be_work_set_rx_mode(struct work_struct *work) 1848 { 1849 struct be_cmd_work *cmd_work = 1850 container_of(work, struct be_cmd_work, work); 1851 1852 __be_set_rx_mode(cmd_work->adapter); 1853 kfree(cmd_work); 1854 } 1855 1856 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 1857 { 1858 struct be_adapter *adapter = netdev_priv(netdev); 1859 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 1860 int status; 1861 1862 if (!sriov_enabled(adapter)) 1863 return -EPERM; 1864 1865 if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs) 1866 return -EINVAL; 1867 1868 /* Proceed further only if user provided MAC is different 1869 * from active MAC 1870 */ 1871 if (ether_addr_equal(mac, vf_cfg->mac_addr)) 1872 return 0; 1873 1874 if (BEx_chip(adapter)) { 1875 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id, 1876 vf + 1); 1877 1878 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle, 1879 &vf_cfg->pmac_id, vf + 1); 1880 } else { 1881 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle, 1882 vf + 1); 1883 } 1884 1885 if (status) { 1886 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x", 1887 mac, vf, status); 1888 return be_cmd_status(status); 1889 } 1890 1891 ether_addr_copy(vf_cfg->mac_addr, mac); 1892 1893 return 0; 1894 } 1895 1896 static int be_get_vf_config(struct net_device *netdev, int vf, 1897 struct ifla_vf_info *vi) 1898 { 1899 struct be_adapter *adapter = netdev_priv(netdev); 1900 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 1901 1902 if (!sriov_enabled(adapter)) 1903 return -EPERM; 1904 1905 if (vf >= adapter->num_vfs) 1906 return -EINVAL; 1907 1908 vi->vf = vf; 1909 vi->max_tx_rate = vf_cfg->tx_rate; 1910 vi->min_tx_rate = 0; 1911 vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK; 1912 vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT; 1913 memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN); 1914 vi->linkstate = adapter->vf_cfg[vf].plink_tracking; 1915 vi->spoofchk = adapter->vf_cfg[vf].spoofchk; 1916 1917 return 0; 1918 } 1919 1920 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan) 1921 { 1922 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 1923 u16 vids[BE_NUM_VLANS_SUPPORTED]; 1924 int vf_if_id = vf_cfg->if_handle; 1925 int status; 1926 1927 /* Enable Transparent VLAN Tagging */ 1928 status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0); 1929 if (status) 1930 return status; 1931 1932 /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */ 1933 vids[0] = 0; 1934 status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1); 1935 if (!status) 1936 dev_info(&adapter->pdev->dev, 1937 "Cleared guest VLANs on VF%d", vf); 1938 1939 /* After TVT is enabled, disallow VFs to program VLAN filters */ 1940 if (vf_cfg->privileges & BE_PRIV_FILTMGMT) { 1941 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges & 1942 ~BE_PRIV_FILTMGMT, vf + 1); 1943 if (!status) 1944 vf_cfg->privileges &= ~BE_PRIV_FILTMGMT; 1945 } 1946 return 0; 1947 } 1948 1949 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf) 1950 { 1951 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 1952 struct device *dev = &adapter->pdev->dev; 1953 int status; 1954 1955 /* Reset Transparent VLAN Tagging. */ 1956 status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1, 1957 vf_cfg->if_handle, 0, 0); 1958 if (status) 1959 return status; 1960 1961 /* Allow VFs to program VLAN filtering */ 1962 if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) { 1963 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges | 1964 BE_PRIV_FILTMGMT, vf + 1); 1965 if (!status) { 1966 vf_cfg->privileges |= BE_PRIV_FILTMGMT; 1967 dev_info(dev, "VF%d: FILTMGMT priv enabled", vf); 1968 } 1969 } 1970 1971 dev_info(dev, 1972 "Disable/re-enable i/f in VM to clear Transparent VLAN tag"); 1973 return 0; 1974 } 1975 1976 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos, 1977 __be16 vlan_proto) 1978 { 1979 struct be_adapter *adapter = netdev_priv(netdev); 1980 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 1981 int status; 1982 1983 if (!sriov_enabled(adapter)) 1984 return -EPERM; 1985 1986 if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7) 1987 return -EINVAL; 1988 1989 if (vlan_proto != htons(ETH_P_8021Q)) 1990 return -EPROTONOSUPPORT; 1991 1992 if (vlan || qos) { 1993 vlan |= qos << VLAN_PRIO_SHIFT; 1994 status = be_set_vf_tvt(adapter, vf, vlan); 1995 } else { 1996 status = be_clear_vf_tvt(adapter, vf); 1997 } 1998 1999 if (status) { 2000 dev_err(&adapter->pdev->dev, 2001 "VLAN %d config on VF %d failed : %#x\n", vlan, vf, 2002 status); 2003 return be_cmd_status(status); 2004 } 2005 2006 vf_cfg->vlan_tag = vlan; 2007 return 0; 2008 } 2009 2010 static int be_set_vf_tx_rate(struct net_device *netdev, int vf, 2011 int min_tx_rate, int max_tx_rate) 2012 { 2013 struct be_adapter *adapter = netdev_priv(netdev); 2014 struct device *dev = &adapter->pdev->dev; 2015 int percent_rate, status = 0; 2016 u16 link_speed = 0; 2017 u8 link_status; 2018 2019 if (!sriov_enabled(adapter)) 2020 return -EPERM; 2021 2022 if (vf >= adapter->num_vfs) 2023 return -EINVAL; 2024 2025 if (min_tx_rate) 2026 return -EINVAL; 2027 2028 if (!max_tx_rate) 2029 goto config_qos; 2030 2031 status = be_cmd_link_status_query(adapter, &link_speed, 2032 &link_status, 0); 2033 if (status) 2034 goto err; 2035 2036 if (!link_status) { 2037 dev_err(dev, "TX-rate setting not allowed when link is down\n"); 2038 status = -ENETDOWN; 2039 goto err; 2040 } 2041 2042 if (max_tx_rate < 100 || max_tx_rate > link_speed) { 2043 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n", 2044 link_speed); 2045 status = -EINVAL; 2046 goto err; 2047 } 2048 2049 /* On Skyhawk the QOS setting must be done only as a % value */ 2050 percent_rate = link_speed / 100; 2051 if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) { 2052 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n", 2053 percent_rate); 2054 status = -EINVAL; 2055 goto err; 2056 } 2057 2058 config_qos: 2059 status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1); 2060 if (status) 2061 goto err; 2062 2063 adapter->vf_cfg[vf].tx_rate = max_tx_rate; 2064 return 0; 2065 2066 err: 2067 dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n", 2068 max_tx_rate, vf); 2069 return be_cmd_status(status); 2070 } 2071 2072 static int be_set_vf_link_state(struct net_device *netdev, int vf, 2073 int link_state) 2074 { 2075 struct be_adapter *adapter = netdev_priv(netdev); 2076 int status; 2077 2078 if (!sriov_enabled(adapter)) 2079 return -EPERM; 2080 2081 if (vf >= adapter->num_vfs) 2082 return -EINVAL; 2083 2084 status = be_cmd_set_logical_link_config(adapter, link_state, vf+1); 2085 if (status) { 2086 dev_err(&adapter->pdev->dev, 2087 "Link state change on VF %d failed: %#x\n", vf, status); 2088 return be_cmd_status(status); 2089 } 2090 2091 adapter->vf_cfg[vf].plink_tracking = link_state; 2092 2093 return 0; 2094 } 2095 2096 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable) 2097 { 2098 struct be_adapter *adapter = netdev_priv(netdev); 2099 struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf]; 2100 u8 spoofchk; 2101 int status; 2102 2103 if (!sriov_enabled(adapter)) 2104 return -EPERM; 2105 2106 if (vf >= adapter->num_vfs) 2107 return -EINVAL; 2108 2109 if (BEx_chip(adapter)) 2110 return -EOPNOTSUPP; 2111 2112 if (enable == vf_cfg->spoofchk) 2113 return 0; 2114 2115 spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK; 2116 2117 status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle, 2118 0, spoofchk); 2119 if (status) { 2120 dev_err(&adapter->pdev->dev, 2121 "Spoofchk change on VF %d failed: %#x\n", vf, status); 2122 return be_cmd_status(status); 2123 } 2124 2125 vf_cfg->spoofchk = enable; 2126 return 0; 2127 } 2128 2129 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts, 2130 ulong now) 2131 { 2132 aic->rx_pkts_prev = rx_pkts; 2133 aic->tx_reqs_prev = tx_pkts; 2134 aic->jiffies = now; 2135 } 2136 2137 static int be_get_new_eqd(struct be_eq_obj *eqo) 2138 { 2139 struct be_adapter *adapter = eqo->adapter; 2140 int eqd, start; 2141 struct be_aic_obj *aic; 2142 struct be_rx_obj *rxo; 2143 struct be_tx_obj *txo; 2144 u64 rx_pkts = 0, tx_pkts = 0, pkts; 2145 ulong now; 2146 u32 pps, delta; 2147 int i; 2148 2149 aic = &adapter->aic_obj[eqo->idx]; 2150 if (!adapter->aic_enabled) { 2151 if (aic->jiffies) 2152 aic->jiffies = 0; 2153 eqd = aic->et_eqd; 2154 return eqd; 2155 } 2156 2157 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) { 2158 do { 2159 start = u64_stats_fetch_begin(&rxo->stats.sync); 2160 pkts = rxo->stats.rx_pkts; 2161 } while (u64_stats_fetch_retry(&rxo->stats.sync, start)); 2162 rx_pkts += pkts; 2163 } 2164 2165 for_all_tx_queues_on_eq(adapter, eqo, txo, i) { 2166 do { 2167 start = u64_stats_fetch_begin(&txo->stats.sync); 2168 pkts = txo->stats.tx_reqs; 2169 } while (u64_stats_fetch_retry(&txo->stats.sync, start)); 2170 tx_pkts += pkts; 2171 } 2172 2173 /* Skip, if wrapped around or first calculation */ 2174 now = jiffies; 2175 if (!aic->jiffies || time_before(now, aic->jiffies) || 2176 rx_pkts < aic->rx_pkts_prev || 2177 tx_pkts < aic->tx_reqs_prev) { 2178 be_aic_update(aic, rx_pkts, tx_pkts, now); 2179 return aic->prev_eqd; 2180 } 2181 2182 delta = jiffies_to_msecs(now - aic->jiffies); 2183 if (delta == 0) 2184 return aic->prev_eqd; 2185 2186 pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) + 2187 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta); 2188 eqd = (pps / 15000) << 2; 2189 2190 if (eqd < 8) 2191 eqd = 0; 2192 eqd = min_t(u32, eqd, aic->max_eqd); 2193 eqd = max_t(u32, eqd, aic->min_eqd); 2194 2195 be_aic_update(aic, rx_pkts, tx_pkts, now); 2196 2197 return eqd; 2198 } 2199 2200 /* For Skyhawk-R only */ 2201 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo) 2202 { 2203 struct be_adapter *adapter = eqo->adapter; 2204 struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx]; 2205 ulong now = jiffies; 2206 int eqd; 2207 u32 mult_enc; 2208 2209 if (!adapter->aic_enabled) 2210 return 0; 2211 2212 if (jiffies_to_msecs(now - aic->jiffies) < 1) 2213 eqd = aic->prev_eqd; 2214 else 2215 eqd = be_get_new_eqd(eqo); 2216 2217 if (eqd > 100) 2218 mult_enc = R2I_DLY_ENC_1; 2219 else if (eqd > 60) 2220 mult_enc = R2I_DLY_ENC_2; 2221 else if (eqd > 20) 2222 mult_enc = R2I_DLY_ENC_3; 2223 else 2224 mult_enc = R2I_DLY_ENC_0; 2225 2226 aic->prev_eqd = eqd; 2227 2228 return mult_enc; 2229 } 2230 2231 void be_eqd_update(struct be_adapter *adapter, bool force_update) 2232 { 2233 struct be_set_eqd set_eqd[MAX_EVT_QS]; 2234 struct be_aic_obj *aic; 2235 struct be_eq_obj *eqo; 2236 int i, num = 0, eqd; 2237 2238 for_all_evt_queues(adapter, eqo, i) { 2239 aic = &adapter->aic_obj[eqo->idx]; 2240 eqd = be_get_new_eqd(eqo); 2241 if (force_update || eqd != aic->prev_eqd) { 2242 set_eqd[num].delay_multiplier = (eqd * 65)/100; 2243 set_eqd[num].eq_id = eqo->q.id; 2244 aic->prev_eqd = eqd; 2245 num++; 2246 } 2247 } 2248 2249 if (num) 2250 be_cmd_modify_eqd(adapter, set_eqd, num); 2251 } 2252 2253 static void be_rx_stats_update(struct be_rx_obj *rxo, 2254 struct be_rx_compl_info *rxcp) 2255 { 2256 struct be_rx_stats *stats = rx_stats(rxo); 2257 2258 u64_stats_update_begin(&stats->sync); 2259 stats->rx_compl++; 2260 stats->rx_bytes += rxcp->pkt_size; 2261 stats->rx_pkts++; 2262 if (rxcp->tunneled) 2263 stats->rx_vxlan_offload_pkts++; 2264 if (rxcp->pkt_type == BE_MULTICAST_PACKET) 2265 stats->rx_mcast_pkts++; 2266 if (rxcp->err) 2267 stats->rx_compl_err++; 2268 u64_stats_update_end(&stats->sync); 2269 } 2270 2271 static inline bool csum_passed(struct be_rx_compl_info *rxcp) 2272 { 2273 /* L4 checksum is not reliable for non TCP/UDP packets. 2274 * Also ignore ipcksm for ipv6 pkts 2275 */ 2276 return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum && 2277 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err; 2278 } 2279 2280 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo) 2281 { 2282 struct be_adapter *adapter = rxo->adapter; 2283 struct be_rx_page_info *rx_page_info; 2284 struct be_queue_info *rxq = &rxo->q; 2285 u32 frag_idx = rxq->tail; 2286 2287 rx_page_info = &rxo->page_info_tbl[frag_idx]; 2288 BUG_ON(!rx_page_info->page); 2289 2290 if (rx_page_info->last_frag) { 2291 dma_unmap_page(&adapter->pdev->dev, 2292 dma_unmap_addr(rx_page_info, bus), 2293 adapter->big_page_size, DMA_FROM_DEVICE); 2294 rx_page_info->last_frag = false; 2295 } else { 2296 dma_sync_single_for_cpu(&adapter->pdev->dev, 2297 dma_unmap_addr(rx_page_info, bus), 2298 rx_frag_size, DMA_FROM_DEVICE); 2299 } 2300 2301 queue_tail_inc(rxq); 2302 atomic_dec(&rxq->used); 2303 return rx_page_info; 2304 } 2305 2306 /* Throwaway the data in the Rx completion */ 2307 static void be_rx_compl_discard(struct be_rx_obj *rxo, 2308 struct be_rx_compl_info *rxcp) 2309 { 2310 struct be_rx_page_info *page_info; 2311 u16 i, num_rcvd = rxcp->num_rcvd; 2312 2313 for (i = 0; i < num_rcvd; i++) { 2314 page_info = get_rx_page_info(rxo); 2315 put_page(page_info->page); 2316 memset(page_info, 0, sizeof(*page_info)); 2317 } 2318 } 2319 2320 /* 2321 * skb_fill_rx_data forms a complete skb for an ether frame 2322 * indicated by rxcp. 2323 */ 2324 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb, 2325 struct be_rx_compl_info *rxcp) 2326 { 2327 struct be_rx_page_info *page_info; 2328 u16 i, j; 2329 u16 hdr_len, curr_frag_len, remaining; 2330 u8 *start; 2331 2332 page_info = get_rx_page_info(rxo); 2333 start = page_address(page_info->page) + page_info->page_offset; 2334 prefetch(start); 2335 2336 /* Copy data in the first descriptor of this completion */ 2337 curr_frag_len = min(rxcp->pkt_size, rx_frag_size); 2338 2339 skb->len = curr_frag_len; 2340 if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */ 2341 memcpy(skb->data, start, curr_frag_len); 2342 /* Complete packet has now been moved to data */ 2343 put_page(page_info->page); 2344 skb->data_len = 0; 2345 skb->tail += curr_frag_len; 2346 } else { 2347 hdr_len = ETH_HLEN; 2348 memcpy(skb->data, start, hdr_len); 2349 skb_shinfo(skb)->nr_frags = 1; 2350 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[0], 2351 page_info->page, 2352 page_info->page_offset + hdr_len, 2353 curr_frag_len - hdr_len); 2354 skb->data_len = curr_frag_len - hdr_len; 2355 skb->truesize += rx_frag_size; 2356 skb->tail += hdr_len; 2357 } 2358 page_info->page = NULL; 2359 2360 if (rxcp->pkt_size <= rx_frag_size) { 2361 BUG_ON(rxcp->num_rcvd != 1); 2362 return; 2363 } 2364 2365 /* More frags present for this completion */ 2366 remaining = rxcp->pkt_size - curr_frag_len; 2367 for (i = 1, j = 0; i < rxcp->num_rcvd; i++) { 2368 page_info = get_rx_page_info(rxo); 2369 curr_frag_len = min(remaining, rx_frag_size); 2370 2371 /* Coalesce all frags from the same physical page in one slot */ 2372 if (page_info->page_offset == 0) { 2373 /* Fresh page */ 2374 j++; 2375 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[j], 2376 page_info->page, 2377 page_info->page_offset, 2378 curr_frag_len); 2379 skb_shinfo(skb)->nr_frags++; 2380 } else { 2381 put_page(page_info->page); 2382 skb_frag_size_add(&skb_shinfo(skb)->frags[j], 2383 curr_frag_len); 2384 } 2385 2386 skb->len += curr_frag_len; 2387 skb->data_len += curr_frag_len; 2388 skb->truesize += rx_frag_size; 2389 remaining -= curr_frag_len; 2390 page_info->page = NULL; 2391 } 2392 BUG_ON(j > MAX_SKB_FRAGS); 2393 } 2394 2395 /* Process the RX completion indicated by rxcp when GRO is disabled */ 2396 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi, 2397 struct be_rx_compl_info *rxcp) 2398 { 2399 struct be_adapter *adapter = rxo->adapter; 2400 struct net_device *netdev = adapter->netdev; 2401 struct sk_buff *skb; 2402 2403 skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE); 2404 if (unlikely(!skb)) { 2405 rx_stats(rxo)->rx_drops_no_skbs++; 2406 be_rx_compl_discard(rxo, rxcp); 2407 return; 2408 } 2409 2410 skb_fill_rx_data(rxo, skb, rxcp); 2411 2412 if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp))) 2413 skb->ip_summed = CHECKSUM_UNNECESSARY; 2414 else 2415 skb_checksum_none_assert(skb); 2416 2417 skb->protocol = eth_type_trans(skb, netdev); 2418 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]); 2419 if (netdev->features & NETIF_F_RXHASH) 2420 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3); 2421 2422 skb->csum_level = rxcp->tunneled; 2423 skb_mark_napi_id(skb, napi); 2424 2425 if (rxcp->vlanf) 2426 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag); 2427 2428 netif_receive_skb(skb); 2429 } 2430 2431 /* Process the RX completion indicated by rxcp when GRO is enabled */ 2432 static void be_rx_compl_process_gro(struct be_rx_obj *rxo, 2433 struct napi_struct *napi, 2434 struct be_rx_compl_info *rxcp) 2435 { 2436 struct be_adapter *adapter = rxo->adapter; 2437 struct be_rx_page_info *page_info; 2438 struct sk_buff *skb = NULL; 2439 u16 remaining, curr_frag_len; 2440 u16 i, j; 2441 2442 skb = napi_get_frags(napi); 2443 if (!skb) { 2444 be_rx_compl_discard(rxo, rxcp); 2445 return; 2446 } 2447 2448 remaining = rxcp->pkt_size; 2449 for (i = 0, j = -1; i < rxcp->num_rcvd; i++) { 2450 page_info = get_rx_page_info(rxo); 2451 2452 curr_frag_len = min(remaining, rx_frag_size); 2453 2454 /* Coalesce all frags from the same physical page in one slot */ 2455 if (i == 0 || page_info->page_offset == 0) { 2456 /* First frag or Fresh page */ 2457 j++; 2458 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[j], 2459 page_info->page, 2460 page_info->page_offset, 2461 curr_frag_len); 2462 } else { 2463 put_page(page_info->page); 2464 skb_frag_size_add(&skb_shinfo(skb)->frags[j], 2465 curr_frag_len); 2466 } 2467 2468 skb->truesize += rx_frag_size; 2469 remaining -= curr_frag_len; 2470 memset(page_info, 0, sizeof(*page_info)); 2471 } 2472 BUG_ON(j > MAX_SKB_FRAGS); 2473 2474 skb_shinfo(skb)->nr_frags = j + 1; 2475 skb->len = rxcp->pkt_size; 2476 skb->data_len = rxcp->pkt_size; 2477 skb->ip_summed = CHECKSUM_UNNECESSARY; 2478 skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]); 2479 if (adapter->netdev->features & NETIF_F_RXHASH) 2480 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3); 2481 2482 skb->csum_level = rxcp->tunneled; 2483 2484 if (rxcp->vlanf) 2485 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag); 2486 2487 napi_gro_frags(napi); 2488 } 2489 2490 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl, 2491 struct be_rx_compl_info *rxcp) 2492 { 2493 rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl); 2494 rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl); 2495 rxcp->err = GET_RX_COMPL_V1_BITS(err, compl); 2496 rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl); 2497 rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl); 2498 rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl); 2499 rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl); 2500 rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl); 2501 rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl); 2502 rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl); 2503 rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl); 2504 if (rxcp->vlanf) { 2505 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl); 2506 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl); 2507 } 2508 rxcp->port = GET_RX_COMPL_V1_BITS(port, compl); 2509 rxcp->tunneled = 2510 GET_RX_COMPL_V1_BITS(tunneled, compl); 2511 } 2512 2513 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl, 2514 struct be_rx_compl_info *rxcp) 2515 { 2516 rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl); 2517 rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl); 2518 rxcp->err = GET_RX_COMPL_V0_BITS(err, compl); 2519 rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl); 2520 rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl); 2521 rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl); 2522 rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl); 2523 rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl); 2524 rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl); 2525 rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl); 2526 rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl); 2527 if (rxcp->vlanf) { 2528 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl); 2529 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl); 2530 } 2531 rxcp->port = GET_RX_COMPL_V0_BITS(port, compl); 2532 rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl); 2533 } 2534 2535 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo) 2536 { 2537 struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq); 2538 struct be_rx_compl_info *rxcp = &rxo->rxcp; 2539 struct be_adapter *adapter = rxo->adapter; 2540 2541 /* For checking the valid bit it is Ok to use either definition as the 2542 * valid bit is at the same position in both v0 and v1 Rx compl */ 2543 if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0) 2544 return NULL; 2545 2546 rmb(); 2547 be_dws_le_to_cpu(compl, sizeof(*compl)); 2548 2549 if (adapter->be3_native) 2550 be_parse_rx_compl_v1(compl, rxcp); 2551 else 2552 be_parse_rx_compl_v0(compl, rxcp); 2553 2554 if (rxcp->ip_frag) 2555 rxcp->l4_csum = 0; 2556 2557 if (rxcp->vlanf) { 2558 /* In QNQ modes, if qnq bit is not set, then the packet was 2559 * tagged only with the transparent outer vlan-tag and must 2560 * not be treated as a vlan packet by host 2561 */ 2562 if (be_is_qnq_mode(adapter) && !rxcp->qnq) 2563 rxcp->vlanf = 0; 2564 2565 if (!lancer_chip(adapter)) 2566 rxcp->vlan_tag = swab16(rxcp->vlan_tag); 2567 2568 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) && 2569 !test_bit(rxcp->vlan_tag, adapter->vids)) 2570 rxcp->vlanf = 0; 2571 } 2572 2573 /* As the compl has been parsed, reset it; we won't touch it again */ 2574 compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0; 2575 2576 queue_tail_inc(&rxo->cq); 2577 return rxcp; 2578 } 2579 2580 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp) 2581 { 2582 u32 order = get_order(size); 2583 2584 if (order > 0) 2585 gfp |= __GFP_COMP; 2586 return alloc_pages(gfp, order); 2587 } 2588 2589 /* 2590 * Allocate a page, split it to fragments of size rx_frag_size and post as 2591 * receive buffers to BE 2592 */ 2593 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed) 2594 { 2595 struct be_adapter *adapter = rxo->adapter; 2596 struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL; 2597 struct be_queue_info *rxq = &rxo->q; 2598 struct page *pagep = NULL; 2599 struct device *dev = &adapter->pdev->dev; 2600 struct be_eth_rx_d *rxd; 2601 u64 page_dmaaddr = 0, frag_dmaaddr; 2602 u32 posted, page_offset = 0, notify = 0; 2603 2604 page_info = &rxo->page_info_tbl[rxq->head]; 2605 for (posted = 0; posted < frags_needed && !page_info->page; posted++) { 2606 if (!pagep) { 2607 pagep = be_alloc_pages(adapter->big_page_size, gfp); 2608 if (unlikely(!pagep)) { 2609 rx_stats(rxo)->rx_post_fail++; 2610 break; 2611 } 2612 page_dmaaddr = dma_map_page(dev, pagep, 0, 2613 adapter->big_page_size, 2614 DMA_FROM_DEVICE); 2615 if (dma_mapping_error(dev, page_dmaaddr)) { 2616 put_page(pagep); 2617 pagep = NULL; 2618 adapter->drv_stats.dma_map_errors++; 2619 break; 2620 } 2621 page_offset = 0; 2622 } else { 2623 get_page(pagep); 2624 page_offset += rx_frag_size; 2625 } 2626 page_info->page_offset = page_offset; 2627 page_info->page = pagep; 2628 2629 rxd = queue_head_node(rxq); 2630 frag_dmaaddr = page_dmaaddr + page_info->page_offset; 2631 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF); 2632 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr)); 2633 2634 /* Any space left in the current big page for another frag? */ 2635 if ((page_offset + rx_frag_size + rx_frag_size) > 2636 adapter->big_page_size) { 2637 pagep = NULL; 2638 page_info->last_frag = true; 2639 dma_unmap_addr_set(page_info, bus, page_dmaaddr); 2640 } else { 2641 dma_unmap_addr_set(page_info, bus, frag_dmaaddr); 2642 } 2643 2644 prev_page_info = page_info; 2645 queue_head_inc(rxq); 2646 page_info = &rxo->page_info_tbl[rxq->head]; 2647 } 2648 2649 /* Mark the last frag of a page when we break out of the above loop 2650 * with no more slots available in the RXQ 2651 */ 2652 if (pagep) { 2653 prev_page_info->last_frag = true; 2654 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr); 2655 } 2656 2657 if (posted) { 2658 atomic_add(posted, &rxq->used); 2659 if (rxo->rx_post_starved) 2660 rxo->rx_post_starved = false; 2661 do { 2662 notify = min(MAX_NUM_POST_ERX_DB, posted); 2663 be_rxq_notify(adapter, rxq->id, notify); 2664 posted -= notify; 2665 } while (posted); 2666 } else if (atomic_read(&rxq->used) == 0) { 2667 /* Let be_worker replenish when memory is available */ 2668 rxo->rx_post_starved = true; 2669 } 2670 } 2671 2672 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status) 2673 { 2674 switch (status) { 2675 case BE_TX_COMP_HDR_PARSE_ERR: 2676 tx_stats(txo)->tx_hdr_parse_err++; 2677 break; 2678 case BE_TX_COMP_NDMA_ERR: 2679 tx_stats(txo)->tx_dma_err++; 2680 break; 2681 case BE_TX_COMP_ACL_ERR: 2682 tx_stats(txo)->tx_spoof_check_err++; 2683 break; 2684 } 2685 } 2686 2687 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status) 2688 { 2689 switch (status) { 2690 case LANCER_TX_COMP_LSO_ERR: 2691 tx_stats(txo)->tx_tso_err++; 2692 break; 2693 case LANCER_TX_COMP_HSW_DROP_MAC_ERR: 2694 case LANCER_TX_COMP_HSW_DROP_VLAN_ERR: 2695 tx_stats(txo)->tx_spoof_check_err++; 2696 break; 2697 case LANCER_TX_COMP_QINQ_ERR: 2698 tx_stats(txo)->tx_qinq_err++; 2699 break; 2700 case LANCER_TX_COMP_PARITY_ERR: 2701 tx_stats(txo)->tx_internal_parity_err++; 2702 break; 2703 case LANCER_TX_COMP_DMA_ERR: 2704 tx_stats(txo)->tx_dma_err++; 2705 break; 2706 case LANCER_TX_COMP_SGE_ERR: 2707 tx_stats(txo)->tx_sge_err++; 2708 break; 2709 } 2710 } 2711 2712 static struct be_tx_compl_info *be_tx_compl_get(struct be_adapter *adapter, 2713 struct be_tx_obj *txo) 2714 { 2715 struct be_queue_info *tx_cq = &txo->cq; 2716 struct be_tx_compl_info *txcp = &txo->txcp; 2717 struct be_eth_tx_compl *compl = queue_tail_node(tx_cq); 2718 2719 if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0) 2720 return NULL; 2721 2722 /* Ensure load ordering of valid bit dword and other dwords below */ 2723 rmb(); 2724 be_dws_le_to_cpu(compl, sizeof(*compl)); 2725 2726 txcp->status = GET_TX_COMPL_BITS(status, compl); 2727 txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl); 2728 2729 if (txcp->status) { 2730 if (lancer_chip(adapter)) { 2731 lancer_update_tx_err(txo, txcp->status); 2732 /* Reset the adapter in case of TSO, 2733 * SGE or Parity error 2734 */ 2735 if (txcp->status == LANCER_TX_COMP_LSO_ERR || 2736 txcp->status == LANCER_TX_COMP_PARITY_ERR || 2737 txcp->status == LANCER_TX_COMP_SGE_ERR) 2738 be_set_error(adapter, BE_ERROR_TX); 2739 } else { 2740 be_update_tx_err(txo, txcp->status); 2741 } 2742 } 2743 2744 if (be_check_error(adapter, BE_ERROR_TX)) 2745 return NULL; 2746 2747 compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0; 2748 queue_tail_inc(tx_cq); 2749 return txcp; 2750 } 2751 2752 static u16 be_tx_compl_process(struct be_adapter *adapter, 2753 struct be_tx_obj *txo, u16 last_index) 2754 { 2755 struct sk_buff **sent_skbs = txo->sent_skb_list; 2756 struct be_queue_info *txq = &txo->q; 2757 struct sk_buff *skb = NULL; 2758 bool unmap_skb_hdr = false; 2759 struct be_eth_wrb *wrb; 2760 u16 num_wrbs = 0; 2761 u32 frag_index; 2762 2763 do { 2764 if (sent_skbs[txq->tail]) { 2765 /* Free skb from prev req */ 2766 if (skb) 2767 dev_consume_skb_any(skb); 2768 skb = sent_skbs[txq->tail]; 2769 sent_skbs[txq->tail] = NULL; 2770 queue_tail_inc(txq); /* skip hdr wrb */ 2771 num_wrbs++; 2772 unmap_skb_hdr = true; 2773 } 2774 wrb = queue_tail_node(txq); 2775 frag_index = txq->tail; 2776 unmap_tx_frag(&adapter->pdev->dev, wrb, 2777 (unmap_skb_hdr && skb_headlen(skb))); 2778 unmap_skb_hdr = false; 2779 queue_tail_inc(txq); 2780 num_wrbs++; 2781 } while (frag_index != last_index); 2782 dev_consume_skb_any(skb); 2783 2784 return num_wrbs; 2785 } 2786 2787 /* Return the number of events in the event queue */ 2788 static inline int events_get(struct be_eq_obj *eqo) 2789 { 2790 struct be_eq_entry *eqe; 2791 int num = 0; 2792 2793 do { 2794 eqe = queue_tail_node(&eqo->q); 2795 if (eqe->evt == 0) 2796 break; 2797 2798 rmb(); 2799 eqe->evt = 0; 2800 num++; 2801 queue_tail_inc(&eqo->q); 2802 } while (true); 2803 2804 return num; 2805 } 2806 2807 /* Leaves the EQ is disarmed state */ 2808 static void be_eq_clean(struct be_eq_obj *eqo) 2809 { 2810 int num = events_get(eqo); 2811 2812 be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0); 2813 } 2814 2815 /* Free posted rx buffers that were not used */ 2816 static void be_rxq_clean(struct be_rx_obj *rxo) 2817 { 2818 struct be_queue_info *rxq = &rxo->q; 2819 struct be_rx_page_info *page_info; 2820 2821 while (atomic_read(&rxq->used) > 0) { 2822 page_info = get_rx_page_info(rxo); 2823 put_page(page_info->page); 2824 memset(page_info, 0, sizeof(*page_info)); 2825 } 2826 BUG_ON(atomic_read(&rxq->used)); 2827 rxq->tail = 0; 2828 rxq->head = 0; 2829 } 2830 2831 static void be_rx_cq_clean(struct be_rx_obj *rxo) 2832 { 2833 struct be_queue_info *rx_cq = &rxo->cq; 2834 struct be_rx_compl_info *rxcp; 2835 struct be_adapter *adapter = rxo->adapter; 2836 int flush_wait = 0; 2837 2838 /* Consume pending rx completions. 2839 * Wait for the flush completion (identified by zero num_rcvd) 2840 * to arrive. Notify CQ even when there are no more CQ entries 2841 * for HW to flush partially coalesced CQ entries. 2842 * In Lancer, there is no need to wait for flush compl. 2843 */ 2844 for (;;) { 2845 rxcp = be_rx_compl_get(rxo); 2846 if (!rxcp) { 2847 if (lancer_chip(adapter)) 2848 break; 2849 2850 if (flush_wait++ > 50 || 2851 be_check_error(adapter, 2852 BE_ERROR_HW)) { 2853 dev_warn(&adapter->pdev->dev, 2854 "did not receive flush compl\n"); 2855 break; 2856 } 2857 be_cq_notify(adapter, rx_cq->id, true, 0); 2858 mdelay(1); 2859 } else { 2860 be_rx_compl_discard(rxo, rxcp); 2861 be_cq_notify(adapter, rx_cq->id, false, 1); 2862 if (rxcp->num_rcvd == 0) 2863 break; 2864 } 2865 } 2866 2867 /* After cleanup, leave the CQ in unarmed state */ 2868 be_cq_notify(adapter, rx_cq->id, false, 0); 2869 } 2870 2871 static void be_tx_compl_clean(struct be_adapter *adapter) 2872 { 2873 struct device *dev = &adapter->pdev->dev; 2874 u16 cmpl = 0, timeo = 0, num_wrbs = 0; 2875 struct be_tx_compl_info *txcp; 2876 struct be_queue_info *txq; 2877 u32 end_idx, notified_idx; 2878 struct be_tx_obj *txo; 2879 int i, pending_txqs; 2880 2881 /* Stop polling for compls when HW has been silent for 10ms */ 2882 do { 2883 pending_txqs = adapter->num_tx_qs; 2884 2885 for_all_tx_queues(adapter, txo, i) { 2886 cmpl = 0; 2887 num_wrbs = 0; 2888 txq = &txo->q; 2889 while ((txcp = be_tx_compl_get(adapter, txo))) { 2890 num_wrbs += 2891 be_tx_compl_process(adapter, txo, 2892 txcp->end_index); 2893 cmpl++; 2894 } 2895 if (cmpl) { 2896 be_cq_notify(adapter, txo->cq.id, false, cmpl); 2897 atomic_sub(num_wrbs, &txq->used); 2898 timeo = 0; 2899 } 2900 if (!be_is_tx_compl_pending(txo)) 2901 pending_txqs--; 2902 } 2903 2904 if (pending_txqs == 0 || ++timeo > 10 || 2905 be_check_error(adapter, BE_ERROR_HW)) 2906 break; 2907 2908 mdelay(1); 2909 } while (true); 2910 2911 /* Free enqueued TX that was never notified to HW */ 2912 for_all_tx_queues(adapter, txo, i) { 2913 txq = &txo->q; 2914 2915 if (atomic_read(&txq->used)) { 2916 dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n", 2917 i, atomic_read(&txq->used)); 2918 notified_idx = txq->tail; 2919 end_idx = txq->tail; 2920 index_adv(&end_idx, atomic_read(&txq->used) - 1, 2921 txq->len); 2922 /* Use the tx-compl process logic to handle requests 2923 * that were not sent to the HW. 2924 */ 2925 num_wrbs = be_tx_compl_process(adapter, txo, end_idx); 2926 atomic_sub(num_wrbs, &txq->used); 2927 BUG_ON(atomic_read(&txq->used)); 2928 txo->pend_wrb_cnt = 0; 2929 /* Since hw was never notified of these requests, 2930 * reset TXQ indices 2931 */ 2932 txq->head = notified_idx; 2933 txq->tail = notified_idx; 2934 } 2935 } 2936 } 2937 2938 static void be_evt_queues_destroy(struct be_adapter *adapter) 2939 { 2940 struct be_eq_obj *eqo; 2941 int i; 2942 2943 for_all_evt_queues(adapter, eqo, i) { 2944 if (eqo->q.created) { 2945 be_eq_clean(eqo); 2946 be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ); 2947 netif_napi_del(&eqo->napi); 2948 free_cpumask_var(eqo->affinity_mask); 2949 } 2950 be_queue_free(adapter, &eqo->q); 2951 } 2952 } 2953 2954 static int be_evt_queues_create(struct be_adapter *adapter) 2955 { 2956 struct be_queue_info *eq; 2957 struct be_eq_obj *eqo; 2958 struct be_aic_obj *aic; 2959 int i, rc; 2960 2961 /* need enough EQs to service both RX and TX queues */ 2962 adapter->num_evt_qs = min_t(u16, num_irqs(adapter), 2963 max(adapter->cfg_num_rx_irqs, 2964 adapter->cfg_num_tx_irqs)); 2965 2966 adapter->aic_enabled = true; 2967 2968 for_all_evt_queues(adapter, eqo, i) { 2969 int numa_node = dev_to_node(&adapter->pdev->dev); 2970 2971 aic = &adapter->aic_obj[i]; 2972 eqo->adapter = adapter; 2973 eqo->idx = i; 2974 aic->max_eqd = BE_MAX_EQD; 2975 2976 eq = &eqo->q; 2977 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN, 2978 sizeof(struct be_eq_entry)); 2979 if (rc) 2980 return rc; 2981 2982 rc = be_cmd_eq_create(adapter, eqo); 2983 if (rc) 2984 return rc; 2985 2986 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL)) 2987 return -ENOMEM; 2988 cpumask_set_cpu(cpumask_local_spread(i, numa_node), 2989 eqo->affinity_mask); 2990 netif_napi_add(adapter->netdev, &eqo->napi, be_poll); 2991 } 2992 return 0; 2993 } 2994 2995 static void be_mcc_queues_destroy(struct be_adapter *adapter) 2996 { 2997 struct be_queue_info *q; 2998 2999 q = &adapter->mcc_obj.q; 3000 if (q->created) 3001 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ); 3002 be_queue_free(adapter, q); 3003 3004 q = &adapter->mcc_obj.cq; 3005 if (q->created) 3006 be_cmd_q_destroy(adapter, q, QTYPE_CQ); 3007 be_queue_free(adapter, q); 3008 } 3009 3010 /* Must be called only after TX qs are created as MCC shares TX EQ */ 3011 static int be_mcc_queues_create(struct be_adapter *adapter) 3012 { 3013 struct be_queue_info *q, *cq; 3014 3015 cq = &adapter->mcc_obj.cq; 3016 if (be_queue_alloc(adapter, cq, MCC_CQ_LEN, 3017 sizeof(struct be_mcc_compl))) 3018 goto err; 3019 3020 /* Use the default EQ for MCC completions */ 3021 if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0)) 3022 goto mcc_cq_free; 3023 3024 q = &adapter->mcc_obj.q; 3025 if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb))) 3026 goto mcc_cq_destroy; 3027 3028 if (be_cmd_mccq_create(adapter, q, cq)) 3029 goto mcc_q_free; 3030 3031 return 0; 3032 3033 mcc_q_free: 3034 be_queue_free(adapter, q); 3035 mcc_cq_destroy: 3036 be_cmd_q_destroy(adapter, cq, QTYPE_CQ); 3037 mcc_cq_free: 3038 be_queue_free(adapter, cq); 3039 err: 3040 return -1; 3041 } 3042 3043 static void be_tx_queues_destroy(struct be_adapter *adapter) 3044 { 3045 struct be_queue_info *q; 3046 struct be_tx_obj *txo; 3047 u8 i; 3048 3049 for_all_tx_queues(adapter, txo, i) { 3050 q = &txo->q; 3051 if (q->created) 3052 be_cmd_q_destroy(adapter, q, QTYPE_TXQ); 3053 be_queue_free(adapter, q); 3054 3055 q = &txo->cq; 3056 if (q->created) 3057 be_cmd_q_destroy(adapter, q, QTYPE_CQ); 3058 be_queue_free(adapter, q); 3059 } 3060 } 3061 3062 static int be_tx_qs_create(struct be_adapter *adapter) 3063 { 3064 struct be_queue_info *cq; 3065 struct be_tx_obj *txo; 3066 struct be_eq_obj *eqo; 3067 int status, i; 3068 3069 adapter->num_tx_qs = min(adapter->num_evt_qs, adapter->cfg_num_tx_irqs); 3070 3071 for_all_tx_queues(adapter, txo, i) { 3072 cq = &txo->cq; 3073 status = be_queue_alloc(adapter, cq, TX_CQ_LEN, 3074 sizeof(struct be_eth_tx_compl)); 3075 if (status) 3076 return status; 3077 3078 u64_stats_init(&txo->stats.sync); 3079 u64_stats_init(&txo->stats.sync_compl); 3080 3081 /* If num_evt_qs is less than num_tx_qs, then more than 3082 * one txq share an eq 3083 */ 3084 eqo = &adapter->eq_obj[i % adapter->num_evt_qs]; 3085 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3); 3086 if (status) 3087 return status; 3088 3089 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN, 3090 sizeof(struct be_eth_wrb)); 3091 if (status) 3092 return status; 3093 3094 status = be_cmd_txq_create(adapter, txo); 3095 if (status) 3096 return status; 3097 3098 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask, 3099 eqo->idx); 3100 } 3101 3102 dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n", 3103 adapter->num_tx_qs); 3104 return 0; 3105 } 3106 3107 static void be_rx_cqs_destroy(struct be_adapter *adapter) 3108 { 3109 struct be_queue_info *q; 3110 struct be_rx_obj *rxo; 3111 int i; 3112 3113 for_all_rx_queues(adapter, rxo, i) { 3114 q = &rxo->cq; 3115 if (q->created) 3116 be_cmd_q_destroy(adapter, q, QTYPE_CQ); 3117 be_queue_free(adapter, q); 3118 } 3119 } 3120 3121 static int be_rx_cqs_create(struct be_adapter *adapter) 3122 { 3123 struct be_queue_info *eq, *cq; 3124 struct be_rx_obj *rxo; 3125 int rc, i; 3126 3127 adapter->num_rss_qs = 3128 min(adapter->num_evt_qs, adapter->cfg_num_rx_irqs); 3129 3130 /* We'll use RSS only if at least 2 RSS rings are supported. */ 3131 if (adapter->num_rss_qs < 2) 3132 adapter->num_rss_qs = 0; 3133 3134 adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq; 3135 3136 /* When the interface is not capable of RSS rings (and there is no 3137 * need to create a default RXQ) we'll still need one RXQ 3138 */ 3139 if (adapter->num_rx_qs == 0) 3140 adapter->num_rx_qs = 1; 3141 3142 adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE; 3143 for_all_rx_queues(adapter, rxo, i) { 3144 rxo->adapter = adapter; 3145 cq = &rxo->cq; 3146 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN, 3147 sizeof(struct be_eth_rx_compl)); 3148 if (rc) 3149 return rc; 3150 3151 u64_stats_init(&rxo->stats.sync); 3152 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q; 3153 rc = be_cmd_cq_create(adapter, cq, eq, false, 3); 3154 if (rc) 3155 return rc; 3156 } 3157 3158 dev_info(&adapter->pdev->dev, 3159 "created %d RX queue(s)\n", adapter->num_rx_qs); 3160 return 0; 3161 } 3162 3163 static irqreturn_t be_intx(int irq, void *dev) 3164 { 3165 struct be_eq_obj *eqo = dev; 3166 struct be_adapter *adapter = eqo->adapter; 3167 int num_evts = 0; 3168 3169 /* IRQ is not expected when NAPI is scheduled as the EQ 3170 * will not be armed. 3171 * But, this can happen on Lancer INTx where it takes 3172 * a while to de-assert INTx or in BE2 where occasionally 3173 * an interrupt may be raised even when EQ is unarmed. 3174 * If NAPI is already scheduled, then counting & notifying 3175 * events will orphan them. 3176 */ 3177 if (napi_schedule_prep(&eqo->napi)) { 3178 num_evts = events_get(eqo); 3179 __napi_schedule(&eqo->napi); 3180 if (num_evts) 3181 eqo->spurious_intr = 0; 3182 } 3183 be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0); 3184 3185 /* Return IRQ_HANDLED only for the first spurious intr 3186 * after a valid intr to stop the kernel from branding 3187 * this irq as a bad one! 3188 */ 3189 if (num_evts || eqo->spurious_intr++ == 0) 3190 return IRQ_HANDLED; 3191 else 3192 return IRQ_NONE; 3193 } 3194 3195 static irqreturn_t be_msix(int irq, void *dev) 3196 { 3197 struct be_eq_obj *eqo = dev; 3198 3199 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0); 3200 napi_schedule(&eqo->napi); 3201 return IRQ_HANDLED; 3202 } 3203 3204 static inline bool do_gro(struct be_rx_compl_info *rxcp) 3205 { 3206 return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false; 3207 } 3208 3209 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi, 3210 int budget) 3211 { 3212 struct be_adapter *adapter = rxo->adapter; 3213 struct be_queue_info *rx_cq = &rxo->cq; 3214 struct be_rx_compl_info *rxcp; 3215 u32 work_done; 3216 u32 frags_consumed = 0; 3217 3218 for (work_done = 0; work_done < budget; work_done++) { 3219 rxcp = be_rx_compl_get(rxo); 3220 if (!rxcp) 3221 break; 3222 3223 /* Is it a flush compl that has no data */ 3224 if (unlikely(rxcp->num_rcvd == 0)) 3225 goto loop_continue; 3226 3227 /* Discard compl with partial DMA Lancer B0 */ 3228 if (unlikely(!rxcp->pkt_size)) { 3229 be_rx_compl_discard(rxo, rxcp); 3230 goto loop_continue; 3231 } 3232 3233 /* On BE drop pkts that arrive due to imperfect filtering in 3234 * promiscuous mode on some skews 3235 */ 3236 if (unlikely(rxcp->port != adapter->port_num && 3237 !lancer_chip(adapter))) { 3238 be_rx_compl_discard(rxo, rxcp); 3239 goto loop_continue; 3240 } 3241 3242 if (do_gro(rxcp)) 3243 be_rx_compl_process_gro(rxo, napi, rxcp); 3244 else 3245 be_rx_compl_process(rxo, napi, rxcp); 3246 3247 loop_continue: 3248 frags_consumed += rxcp->num_rcvd; 3249 be_rx_stats_update(rxo, rxcp); 3250 } 3251 3252 if (work_done) { 3253 be_cq_notify(adapter, rx_cq->id, true, work_done); 3254 3255 /* When an rx-obj gets into post_starved state, just 3256 * let be_worker do the posting. 3257 */ 3258 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM && 3259 !rxo->rx_post_starved) 3260 be_post_rx_frags(rxo, GFP_ATOMIC, 3261 max_t(u32, MAX_RX_POST, 3262 frags_consumed)); 3263 } 3264 3265 return work_done; 3266 } 3267 3268 3269 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo, 3270 int idx) 3271 { 3272 int num_wrbs = 0, work_done = 0; 3273 struct be_tx_compl_info *txcp; 3274 3275 while ((txcp = be_tx_compl_get(adapter, txo))) { 3276 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index); 3277 work_done++; 3278 } 3279 3280 if (work_done) { 3281 be_cq_notify(adapter, txo->cq.id, true, work_done); 3282 atomic_sub(num_wrbs, &txo->q.used); 3283 3284 /* As Tx wrbs have been freed up, wake up netdev queue 3285 * if it was stopped due to lack of tx wrbs. */ 3286 if (__netif_subqueue_stopped(adapter->netdev, idx) && 3287 be_can_txq_wake(txo)) { 3288 netif_wake_subqueue(adapter->netdev, idx); 3289 } 3290 3291 u64_stats_update_begin(&tx_stats(txo)->sync_compl); 3292 tx_stats(txo)->tx_compl += work_done; 3293 u64_stats_update_end(&tx_stats(txo)->sync_compl); 3294 } 3295 } 3296 3297 int be_poll(struct napi_struct *napi, int budget) 3298 { 3299 struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi); 3300 struct be_adapter *adapter = eqo->adapter; 3301 int max_work = 0, work, i, num_evts; 3302 struct be_rx_obj *rxo; 3303 struct be_tx_obj *txo; 3304 u32 mult_enc = 0; 3305 3306 num_evts = events_get(eqo); 3307 3308 for_all_tx_queues_on_eq(adapter, eqo, txo, i) 3309 be_process_tx(adapter, txo, i); 3310 3311 /* This loop will iterate twice for EQ0 in which 3312 * completions of the last RXQ (default one) are also processed 3313 * For other EQs the loop iterates only once 3314 */ 3315 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) { 3316 work = be_process_rx(rxo, napi, budget); 3317 max_work = max(work, max_work); 3318 } 3319 3320 if (is_mcc_eqo(eqo)) 3321 be_process_mcc(adapter); 3322 3323 if (max_work < budget) { 3324 napi_complete_done(napi, max_work); 3325 3326 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt 3327 * delay via a delay multiplier encoding value 3328 */ 3329 if (skyhawk_chip(adapter)) 3330 mult_enc = be_get_eq_delay_mult_enc(eqo); 3331 3332 be_eq_notify(adapter, eqo->q.id, true, false, num_evts, 3333 mult_enc); 3334 } else { 3335 /* As we'll continue in polling mode, count and clear events */ 3336 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0); 3337 } 3338 return max_work; 3339 } 3340 3341 void be_detect_error(struct be_adapter *adapter) 3342 { 3343 u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0; 3344 u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0; 3345 struct device *dev = &adapter->pdev->dev; 3346 u16 val; 3347 u32 i; 3348 3349 if (be_check_error(adapter, BE_ERROR_HW)) 3350 return; 3351 3352 if (lancer_chip(adapter)) { 3353 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET); 3354 if (sliport_status & SLIPORT_STATUS_ERR_MASK) { 3355 be_set_error(adapter, BE_ERROR_UE); 3356 sliport_err1 = ioread32(adapter->db + 3357 SLIPORT_ERROR1_OFFSET); 3358 sliport_err2 = ioread32(adapter->db + 3359 SLIPORT_ERROR2_OFFSET); 3360 /* Do not log error messages if its a FW reset */ 3361 if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 && 3362 sliport_err2 == SLIPORT_ERROR_FW_RESET2) { 3363 dev_info(dev, "Reset is in progress\n"); 3364 } else { 3365 dev_err(dev, "Error detected in the card\n"); 3366 dev_err(dev, "ERR: sliport status 0x%x\n", 3367 sliport_status); 3368 dev_err(dev, "ERR: sliport error1 0x%x\n", 3369 sliport_err1); 3370 dev_err(dev, "ERR: sliport error2 0x%x\n", 3371 sliport_err2); 3372 } 3373 } 3374 } else { 3375 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW); 3376 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH); 3377 ue_lo_mask = ioread32(adapter->pcicfg + 3378 PCICFG_UE_STATUS_LOW_MASK); 3379 ue_hi_mask = ioread32(adapter->pcicfg + 3380 PCICFG_UE_STATUS_HI_MASK); 3381 3382 ue_lo = (ue_lo & ~ue_lo_mask); 3383 ue_hi = (ue_hi & ~ue_hi_mask); 3384 3385 if (ue_lo || ue_hi) { 3386 /* On certain platforms BE3 hardware can indicate 3387 * spurious UEs. In case of a UE in the chip, 3388 * the POST register correctly reports either a 3389 * FAT_LOG_START state (FW is currently dumping 3390 * FAT log data) or a ARMFW_UE state. Check for the 3391 * above states to ascertain if the UE is valid or not. 3392 */ 3393 if (BE3_chip(adapter)) { 3394 val = be_POST_stage_get(adapter); 3395 if ((val & POST_STAGE_FAT_LOG_START) 3396 != POST_STAGE_FAT_LOG_START && 3397 (val & POST_STAGE_ARMFW_UE) 3398 != POST_STAGE_ARMFW_UE && 3399 (val & POST_STAGE_RECOVERABLE_ERR) 3400 != POST_STAGE_RECOVERABLE_ERR) 3401 return; 3402 } 3403 3404 dev_err(dev, "Error detected in the adapter"); 3405 be_set_error(adapter, BE_ERROR_UE); 3406 3407 for (i = 0; ue_lo; ue_lo >>= 1, i++) { 3408 if (ue_lo & 1) 3409 dev_err(dev, "UE: %s bit set\n", 3410 ue_status_low_desc[i]); 3411 } 3412 for (i = 0; ue_hi; ue_hi >>= 1, i++) { 3413 if (ue_hi & 1) 3414 dev_err(dev, "UE: %s bit set\n", 3415 ue_status_hi_desc[i]); 3416 } 3417 } 3418 } 3419 } 3420 3421 static void be_msix_disable(struct be_adapter *adapter) 3422 { 3423 if (msix_enabled(adapter)) { 3424 pci_disable_msix(adapter->pdev); 3425 adapter->num_msix_vec = 0; 3426 adapter->num_msix_roce_vec = 0; 3427 } 3428 } 3429 3430 static int be_msix_enable(struct be_adapter *adapter) 3431 { 3432 unsigned int i, max_roce_eqs; 3433 struct device *dev = &adapter->pdev->dev; 3434 int num_vec; 3435 3436 /* If RoCE is supported, program the max number of vectors that 3437 * could be used for NIC and RoCE, else, just program the number 3438 * we'll use initially. 3439 */ 3440 if (be_roce_supported(adapter)) { 3441 max_roce_eqs = 3442 be_max_func_eqs(adapter) - be_max_nic_eqs(adapter); 3443 max_roce_eqs = min(max_roce_eqs, num_online_cpus()); 3444 num_vec = be_max_any_irqs(adapter) + max_roce_eqs; 3445 } else { 3446 num_vec = max(adapter->cfg_num_rx_irqs, 3447 adapter->cfg_num_tx_irqs); 3448 } 3449 3450 for (i = 0; i < num_vec; i++) 3451 adapter->msix_entries[i].entry = i; 3452 3453 num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries, 3454 MIN_MSIX_VECTORS, num_vec); 3455 if (num_vec < 0) 3456 goto fail; 3457 3458 if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) { 3459 adapter->num_msix_roce_vec = num_vec / 2; 3460 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n", 3461 adapter->num_msix_roce_vec); 3462 } 3463 3464 adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec; 3465 3466 dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n", 3467 adapter->num_msix_vec); 3468 return 0; 3469 3470 fail: 3471 dev_warn(dev, "MSIx enable failed\n"); 3472 3473 /* INTx is not supported in VFs, so fail probe if enable_msix fails */ 3474 if (be_virtfn(adapter)) 3475 return num_vec; 3476 return 0; 3477 } 3478 3479 static inline int be_msix_vec_get(struct be_adapter *adapter, 3480 struct be_eq_obj *eqo) 3481 { 3482 return adapter->msix_entries[eqo->msix_idx].vector; 3483 } 3484 3485 static int be_msix_register(struct be_adapter *adapter) 3486 { 3487 struct net_device *netdev = adapter->netdev; 3488 struct be_eq_obj *eqo; 3489 int status, i, vec; 3490 3491 for_all_evt_queues(adapter, eqo, i) { 3492 sprintf(eqo->desc, "%s-q%d", netdev->name, i); 3493 vec = be_msix_vec_get(adapter, eqo); 3494 status = request_irq(vec, be_msix, 0, eqo->desc, eqo); 3495 if (status) 3496 goto err_msix; 3497 3498 irq_update_affinity_hint(vec, eqo->affinity_mask); 3499 } 3500 3501 return 0; 3502 err_msix: 3503 for (i--; i >= 0; i--) { 3504 eqo = &adapter->eq_obj[i]; 3505 free_irq(be_msix_vec_get(adapter, eqo), eqo); 3506 } 3507 dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n", 3508 status); 3509 be_msix_disable(adapter); 3510 return status; 3511 } 3512 3513 static int be_irq_register(struct be_adapter *adapter) 3514 { 3515 struct net_device *netdev = adapter->netdev; 3516 int status; 3517 3518 if (msix_enabled(adapter)) { 3519 status = be_msix_register(adapter); 3520 if (status == 0) 3521 goto done; 3522 /* INTx is not supported for VF */ 3523 if (be_virtfn(adapter)) 3524 return status; 3525 } 3526 3527 /* INTx: only the first EQ is used */ 3528 netdev->irq = adapter->pdev->irq; 3529 status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name, 3530 &adapter->eq_obj[0]); 3531 if (status) { 3532 dev_err(&adapter->pdev->dev, 3533 "INTx request IRQ failed - err %d\n", status); 3534 return status; 3535 } 3536 done: 3537 adapter->isr_registered = true; 3538 return 0; 3539 } 3540 3541 static void be_irq_unregister(struct be_adapter *adapter) 3542 { 3543 struct net_device *netdev = adapter->netdev; 3544 struct be_eq_obj *eqo; 3545 int i, vec; 3546 3547 if (!adapter->isr_registered) 3548 return; 3549 3550 /* INTx */ 3551 if (!msix_enabled(adapter)) { 3552 free_irq(netdev->irq, &adapter->eq_obj[0]); 3553 goto done; 3554 } 3555 3556 /* MSIx */ 3557 for_all_evt_queues(adapter, eqo, i) { 3558 vec = be_msix_vec_get(adapter, eqo); 3559 irq_update_affinity_hint(vec, NULL); 3560 free_irq(vec, eqo); 3561 } 3562 3563 done: 3564 adapter->isr_registered = false; 3565 } 3566 3567 static void be_rx_qs_destroy(struct be_adapter *adapter) 3568 { 3569 struct rss_info *rss = &adapter->rss_info; 3570 struct be_queue_info *q; 3571 struct be_rx_obj *rxo; 3572 int i; 3573 3574 for_all_rx_queues(adapter, rxo, i) { 3575 q = &rxo->q; 3576 if (q->created) { 3577 /* If RXQs are destroyed while in an "out of buffer" 3578 * state, there is a possibility of an HW stall on 3579 * Lancer. So, post 64 buffers to each queue to relieve 3580 * the "out of buffer" condition. 3581 * Make sure there's space in the RXQ before posting. 3582 */ 3583 if (lancer_chip(adapter)) { 3584 be_rx_cq_clean(rxo); 3585 if (atomic_read(&q->used) == 0) 3586 be_post_rx_frags(rxo, GFP_KERNEL, 3587 MAX_RX_POST); 3588 } 3589 3590 be_cmd_rxq_destroy(adapter, q); 3591 be_rx_cq_clean(rxo); 3592 be_rxq_clean(rxo); 3593 } 3594 be_queue_free(adapter, q); 3595 } 3596 3597 if (rss->rss_flags) { 3598 rss->rss_flags = RSS_ENABLE_NONE; 3599 be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags, 3600 128, rss->rss_hkey); 3601 } 3602 } 3603 3604 static void be_disable_if_filters(struct be_adapter *adapter) 3605 { 3606 /* Don't delete MAC on BE3 VFs without FILTMGMT privilege */ 3607 if (!BEx_chip(adapter) || !be_virtfn(adapter) || 3608 check_privilege(adapter, BE_PRIV_FILTMGMT)) { 3609 be_dev_mac_del(adapter, adapter->pmac_id[0]); 3610 eth_zero_addr(adapter->dev_mac); 3611 } 3612 3613 be_clear_uc_list(adapter); 3614 be_clear_mc_list(adapter); 3615 3616 /* The IFACE flags are enabled in the open path and cleared 3617 * in the close path. When a VF gets detached from the host and 3618 * assigned to a VM the following happens: 3619 * - VF's IFACE flags get cleared in the detach path 3620 * - IFACE create is issued by the VF in the attach path 3621 * Due to a bug in the BE3/Skyhawk-R FW 3622 * (Lancer FW doesn't have the bug), the IFACE capability flags 3623 * specified along with the IFACE create cmd issued by a VF are not 3624 * honoured by FW. As a consequence, if a *new* driver 3625 * (that enables/disables IFACE flags in open/close) 3626 * is loaded in the host and an *old* driver is * used by a VM/VF, 3627 * the IFACE gets created *without* the needed flags. 3628 * To avoid this, disable RX-filter flags only for Lancer. 3629 */ 3630 if (lancer_chip(adapter)) { 3631 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF); 3632 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS; 3633 } 3634 } 3635 3636 static int be_close(struct net_device *netdev) 3637 { 3638 struct be_adapter *adapter = netdev_priv(netdev); 3639 struct be_eq_obj *eqo; 3640 int i; 3641 3642 /* This protection is needed as be_close() may be called even when the 3643 * adapter is in cleared state (after eeh perm failure) 3644 */ 3645 if (!(adapter->flags & BE_FLAGS_SETUP_DONE)) 3646 return 0; 3647 3648 /* Before attempting cleanup ensure all the pending cmds in the 3649 * config_wq have finished execution 3650 */ 3651 flush_workqueue(be_wq); 3652 3653 be_disable_if_filters(adapter); 3654 3655 if (adapter->flags & BE_FLAGS_NAPI_ENABLED) { 3656 for_all_evt_queues(adapter, eqo, i) { 3657 napi_disable(&eqo->napi); 3658 } 3659 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED; 3660 } 3661 3662 be_async_mcc_disable(adapter); 3663 3664 /* Wait for all pending tx completions to arrive so that 3665 * all tx skbs are freed. 3666 */ 3667 netif_tx_disable(netdev); 3668 be_tx_compl_clean(adapter); 3669 3670 be_rx_qs_destroy(adapter); 3671 3672 for_all_evt_queues(adapter, eqo, i) { 3673 if (msix_enabled(adapter)) 3674 synchronize_irq(be_msix_vec_get(adapter, eqo)); 3675 else 3676 synchronize_irq(netdev->irq); 3677 be_eq_clean(eqo); 3678 } 3679 3680 be_irq_unregister(adapter); 3681 3682 return 0; 3683 } 3684 3685 static int be_rx_qs_create(struct be_adapter *adapter) 3686 { 3687 struct rss_info *rss = &adapter->rss_info; 3688 u8 rss_key[RSS_HASH_KEY_LEN]; 3689 struct be_rx_obj *rxo; 3690 int rc, i, j; 3691 3692 for_all_rx_queues(adapter, rxo, i) { 3693 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN, 3694 sizeof(struct be_eth_rx_d)); 3695 if (rc) 3696 return rc; 3697 } 3698 3699 if (adapter->need_def_rxq || !adapter->num_rss_qs) { 3700 rxo = default_rxo(adapter); 3701 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id, 3702 rx_frag_size, adapter->if_handle, 3703 false, &rxo->rss_id); 3704 if (rc) 3705 return rc; 3706 } 3707 3708 for_all_rss_queues(adapter, rxo, i) { 3709 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id, 3710 rx_frag_size, adapter->if_handle, 3711 true, &rxo->rss_id); 3712 if (rc) 3713 return rc; 3714 } 3715 3716 if (be_multi_rxq(adapter)) { 3717 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) { 3718 for_all_rss_queues(adapter, rxo, i) { 3719 if ((j + i) >= RSS_INDIR_TABLE_LEN) 3720 break; 3721 rss->rsstable[j + i] = rxo->rss_id; 3722 rss->rss_queue[j + i] = i; 3723 } 3724 } 3725 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 | 3726 RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6; 3727 3728 if (!BEx_chip(adapter)) 3729 rss->rss_flags |= RSS_ENABLE_UDP_IPV4 | 3730 RSS_ENABLE_UDP_IPV6; 3731 3732 netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN); 3733 rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags, 3734 RSS_INDIR_TABLE_LEN, rss_key); 3735 if (rc) { 3736 rss->rss_flags = RSS_ENABLE_NONE; 3737 return rc; 3738 } 3739 3740 memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN); 3741 } else { 3742 /* Disable RSS, if only default RX Q is created */ 3743 rss->rss_flags = RSS_ENABLE_NONE; 3744 } 3745 3746 3747 /* Post 1 less than RXQ-len to avoid head being equal to tail, 3748 * which is a queue empty condition 3749 */ 3750 for_all_rx_queues(adapter, rxo, i) 3751 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1); 3752 3753 return 0; 3754 } 3755 3756 static int be_enable_if_filters(struct be_adapter *adapter) 3757 { 3758 int status; 3759 3760 status = be_cmd_rx_filter(adapter, BE_IF_FILT_FLAGS_BASIC, ON); 3761 if (status) 3762 return status; 3763 3764 /* Normally this condition usually true as the ->dev_mac is zeroed. 3765 * But on BE3 VFs the initial MAC is pre-programmed by PF and 3766 * subsequent be_dev_mac_add() can fail (after fresh boot) 3767 */ 3768 if (!ether_addr_equal(adapter->dev_mac, adapter->netdev->dev_addr)) { 3769 int old_pmac_id = -1; 3770 3771 /* Remember old programmed MAC if any - can happen on BE3 VF */ 3772 if (!is_zero_ether_addr(adapter->dev_mac)) 3773 old_pmac_id = adapter->pmac_id[0]; 3774 3775 status = be_dev_mac_add(adapter, adapter->netdev->dev_addr); 3776 if (status) 3777 return status; 3778 3779 /* Delete the old programmed MAC as we successfully programmed 3780 * a new MAC 3781 */ 3782 if (old_pmac_id >= 0 && old_pmac_id != adapter->pmac_id[0]) 3783 be_dev_mac_del(adapter, old_pmac_id); 3784 3785 ether_addr_copy(adapter->dev_mac, adapter->netdev->dev_addr); 3786 } 3787 3788 if (adapter->vlans_added) 3789 be_vid_config(adapter); 3790 3791 __be_set_rx_mode(adapter); 3792 3793 return 0; 3794 } 3795 3796 static int be_open(struct net_device *netdev) 3797 { 3798 struct be_adapter *adapter = netdev_priv(netdev); 3799 struct be_eq_obj *eqo; 3800 struct be_rx_obj *rxo; 3801 struct be_tx_obj *txo; 3802 u8 link_status; 3803 int status, i; 3804 3805 status = be_rx_qs_create(adapter); 3806 if (status) 3807 goto err; 3808 3809 status = be_enable_if_filters(adapter); 3810 if (status) 3811 goto err; 3812 3813 status = be_irq_register(adapter); 3814 if (status) 3815 goto err; 3816 3817 for_all_rx_queues(adapter, rxo, i) 3818 be_cq_notify(adapter, rxo->cq.id, true, 0); 3819 3820 for_all_tx_queues(adapter, txo, i) 3821 be_cq_notify(adapter, txo->cq.id, true, 0); 3822 3823 be_async_mcc_enable(adapter); 3824 3825 for_all_evt_queues(adapter, eqo, i) { 3826 napi_enable(&eqo->napi); 3827 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0); 3828 } 3829 adapter->flags |= BE_FLAGS_NAPI_ENABLED; 3830 3831 status = be_cmd_link_status_query(adapter, NULL, &link_status, 0); 3832 if (!status) 3833 be_link_status_update(adapter, link_status); 3834 3835 netif_tx_start_all_queues(netdev); 3836 3837 udp_tunnel_nic_reset_ntf(netdev); 3838 3839 return 0; 3840 err: 3841 be_close(adapter->netdev); 3842 return -EIO; 3843 } 3844 3845 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac) 3846 { 3847 u32 addr; 3848 3849 addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0); 3850 3851 mac[5] = (u8)(addr & 0xFF); 3852 mac[4] = (u8)((addr >> 8) & 0xFF); 3853 mac[3] = (u8)((addr >> 16) & 0xFF); 3854 /* Use the OUI from the current MAC address */ 3855 memcpy(mac, adapter->netdev->dev_addr, 3); 3856 } 3857 3858 /* 3859 * Generate a seed MAC address from the PF MAC Address using jhash. 3860 * MAC Address for VFs are assigned incrementally starting from the seed. 3861 * These addresses are programmed in the ASIC by the PF and the VF driver 3862 * queries for the MAC address during its probe. 3863 */ 3864 static int be_vf_eth_addr_config(struct be_adapter *adapter) 3865 { 3866 u32 vf; 3867 int status = 0; 3868 u8 mac[ETH_ALEN]; 3869 struct be_vf_cfg *vf_cfg; 3870 3871 be_vf_eth_addr_generate(adapter, mac); 3872 3873 for_all_vfs(adapter, vf_cfg, vf) { 3874 if (BEx_chip(adapter)) 3875 status = be_cmd_pmac_add(adapter, mac, 3876 vf_cfg->if_handle, 3877 &vf_cfg->pmac_id, vf + 1); 3878 else 3879 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle, 3880 vf + 1); 3881 3882 if (status) 3883 dev_err(&adapter->pdev->dev, 3884 "Mac address assignment failed for VF %d\n", 3885 vf); 3886 else 3887 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN); 3888 3889 mac[5] += 1; 3890 } 3891 return status; 3892 } 3893 3894 static int be_vfs_mac_query(struct be_adapter *adapter) 3895 { 3896 int status, vf; 3897 u8 mac[ETH_ALEN]; 3898 struct be_vf_cfg *vf_cfg; 3899 3900 for_all_vfs(adapter, vf_cfg, vf) { 3901 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id, 3902 mac, vf_cfg->if_handle, 3903 false, vf+1); 3904 if (status) 3905 return status; 3906 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN); 3907 } 3908 return 0; 3909 } 3910 3911 static void be_vf_clear(struct be_adapter *adapter) 3912 { 3913 struct be_vf_cfg *vf_cfg; 3914 u32 vf; 3915 3916 if (pci_vfs_assigned(adapter->pdev)) { 3917 dev_warn(&adapter->pdev->dev, 3918 "VFs are assigned to VMs: not disabling VFs\n"); 3919 goto done; 3920 } 3921 3922 pci_disable_sriov(adapter->pdev); 3923 3924 for_all_vfs(adapter, vf_cfg, vf) { 3925 if (BEx_chip(adapter)) 3926 be_cmd_pmac_del(adapter, vf_cfg->if_handle, 3927 vf_cfg->pmac_id, vf + 1); 3928 else 3929 be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle, 3930 vf + 1); 3931 3932 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1); 3933 } 3934 3935 if (BE3_chip(adapter)) 3936 be_cmd_set_hsw_config(adapter, 0, 0, 3937 adapter->if_handle, 3938 PORT_FWD_TYPE_PASSTHRU, 0); 3939 done: 3940 kfree(adapter->vf_cfg); 3941 adapter->num_vfs = 0; 3942 adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED; 3943 } 3944 3945 static void be_clear_queues(struct be_adapter *adapter) 3946 { 3947 be_mcc_queues_destroy(adapter); 3948 be_rx_cqs_destroy(adapter); 3949 be_tx_queues_destroy(adapter); 3950 be_evt_queues_destroy(adapter); 3951 } 3952 3953 static void be_cancel_worker(struct be_adapter *adapter) 3954 { 3955 if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) { 3956 cancel_delayed_work_sync(&adapter->work); 3957 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED; 3958 } 3959 } 3960 3961 static void be_cancel_err_detection(struct be_adapter *adapter) 3962 { 3963 struct be_error_recovery *err_rec = &adapter->error_recovery; 3964 3965 if (!be_err_recovery_workq) 3966 return; 3967 3968 if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) { 3969 cancel_delayed_work_sync(&err_rec->err_detection_work); 3970 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED; 3971 } 3972 } 3973 3974 /* VxLAN offload Notes: 3975 * 3976 * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't 3977 * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload 3978 * is expected to work across all types of IP tunnels once exported. Skyhawk 3979 * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN 3980 * offloads in hw_enc_features only when a VxLAN port is added. If other (non 3981 * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for 3982 * those other tunnels are unexported on the fly through ndo_features_check(). 3983 */ 3984 static int be_vxlan_set_port(struct net_device *netdev, unsigned int table, 3985 unsigned int entry, struct udp_tunnel_info *ti) 3986 { 3987 struct be_adapter *adapter = netdev_priv(netdev); 3988 struct device *dev = &adapter->pdev->dev; 3989 int status; 3990 3991 status = be_cmd_manage_iface(adapter, adapter->if_handle, 3992 OP_CONVERT_NORMAL_TO_TUNNEL); 3993 if (status) { 3994 dev_warn(dev, "Failed to convert normal interface to tunnel\n"); 3995 return status; 3996 } 3997 adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS; 3998 3999 status = be_cmd_set_vxlan_port(adapter, ti->port); 4000 if (status) { 4001 dev_warn(dev, "Failed to add VxLAN port\n"); 4002 return status; 4003 } 4004 adapter->vxlan_port = ti->port; 4005 4006 netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | 4007 NETIF_F_TSO | NETIF_F_TSO6 | 4008 NETIF_F_GSO_UDP_TUNNEL; 4009 4010 dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n", 4011 be16_to_cpu(ti->port)); 4012 return 0; 4013 } 4014 4015 static int be_vxlan_unset_port(struct net_device *netdev, unsigned int table, 4016 unsigned int entry, struct udp_tunnel_info *ti) 4017 { 4018 struct be_adapter *adapter = netdev_priv(netdev); 4019 4020 if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) 4021 be_cmd_manage_iface(adapter, adapter->if_handle, 4022 OP_CONVERT_TUNNEL_TO_NORMAL); 4023 4024 if (adapter->vxlan_port) 4025 be_cmd_set_vxlan_port(adapter, 0); 4026 4027 adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS; 4028 adapter->vxlan_port = 0; 4029 4030 netdev->hw_enc_features = 0; 4031 return 0; 4032 } 4033 4034 static const struct udp_tunnel_nic_info be_udp_tunnels = { 4035 .set_port = be_vxlan_set_port, 4036 .unset_port = be_vxlan_unset_port, 4037 .flags = UDP_TUNNEL_NIC_INFO_OPEN_ONLY, 4038 .tables = { 4039 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, }, 4040 }, 4041 }; 4042 4043 static void be_calculate_vf_res(struct be_adapter *adapter, u16 num_vfs, 4044 struct be_resources *vft_res) 4045 { 4046 struct be_resources res = adapter->pool_res; 4047 u32 vf_if_cap_flags = res.vf_if_cap_flags; 4048 struct be_resources res_mod = {0}; 4049 u16 num_vf_qs = 1; 4050 4051 /* Distribute the queue resources among the PF and it's VFs */ 4052 if (num_vfs) { 4053 /* Divide the rx queues evenly among the VFs and the PF, capped 4054 * at VF-EQ-count. Any remainder queues belong to the PF. 4055 */ 4056 num_vf_qs = min(SH_VF_MAX_NIC_EQS, 4057 res.max_rss_qs / (num_vfs + 1)); 4058 4059 /* Skyhawk-R chip supports only MAX_PORT_RSS_TABLES 4060 * RSS Tables per port. Provide RSS on VFs, only if number of 4061 * VFs requested is less than it's PF Pool's RSS Tables limit. 4062 */ 4063 if (num_vfs >= be_max_pf_pool_rss_tables(adapter)) 4064 num_vf_qs = 1; 4065 } 4066 4067 /* Resource with fields set to all '1's by GET_PROFILE_CONFIG cmd, 4068 * which are modifiable using SET_PROFILE_CONFIG cmd. 4069 */ 4070 be_cmd_get_profile_config(adapter, &res_mod, NULL, ACTIVE_PROFILE_TYPE, 4071 RESOURCE_MODIFIABLE, 0); 4072 4073 /* If RSS IFACE capability flags are modifiable for a VF, set the 4074 * capability flag as valid and set RSS and DEFQ_RSS IFACE flags if 4075 * more than 1 RSSQ is available for a VF. 4076 * Otherwise, provision only 1 queue pair for VF. 4077 */ 4078 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_RSS) { 4079 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT); 4080 if (num_vf_qs > 1) { 4081 vf_if_cap_flags |= BE_IF_FLAGS_RSS; 4082 if (res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS) 4083 vf_if_cap_flags |= BE_IF_FLAGS_DEFQ_RSS; 4084 } else { 4085 vf_if_cap_flags &= ~(BE_IF_FLAGS_RSS | 4086 BE_IF_FLAGS_DEFQ_RSS); 4087 } 4088 } else { 4089 num_vf_qs = 1; 4090 } 4091 4092 if (res_mod.vf_if_cap_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) { 4093 vft_res->flags |= BIT(IF_CAPS_FLAGS_VALID_SHIFT); 4094 vf_if_cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS; 4095 } 4096 4097 vft_res->vf_if_cap_flags = vf_if_cap_flags; 4098 vft_res->max_rx_qs = num_vf_qs; 4099 vft_res->max_rss_qs = num_vf_qs; 4100 vft_res->max_tx_qs = res.max_tx_qs / (num_vfs + 1); 4101 vft_res->max_cq_count = res.max_cq_count / (num_vfs + 1); 4102 4103 /* Distribute unicast MACs, VLANs, IFACE count and MCCQ count equally 4104 * among the PF and it's VFs, if the fields are changeable 4105 */ 4106 if (res_mod.max_uc_mac == FIELD_MODIFIABLE) 4107 vft_res->max_uc_mac = res.max_uc_mac / (num_vfs + 1); 4108 4109 if (res_mod.max_vlans == FIELD_MODIFIABLE) 4110 vft_res->max_vlans = res.max_vlans / (num_vfs + 1); 4111 4112 if (res_mod.max_iface_count == FIELD_MODIFIABLE) 4113 vft_res->max_iface_count = res.max_iface_count / (num_vfs + 1); 4114 4115 if (res_mod.max_mcc_count == FIELD_MODIFIABLE) 4116 vft_res->max_mcc_count = res.max_mcc_count / (num_vfs + 1); 4117 } 4118 4119 static void be_if_destroy(struct be_adapter *adapter) 4120 { 4121 be_cmd_if_destroy(adapter, adapter->if_handle, 0); 4122 4123 kfree(adapter->pmac_id); 4124 adapter->pmac_id = NULL; 4125 4126 kfree(adapter->mc_list); 4127 adapter->mc_list = NULL; 4128 4129 kfree(adapter->uc_list); 4130 adapter->uc_list = NULL; 4131 } 4132 4133 static int be_clear(struct be_adapter *adapter) 4134 { 4135 struct pci_dev *pdev = adapter->pdev; 4136 struct be_resources vft_res = {0}; 4137 4138 be_cancel_worker(adapter); 4139 4140 flush_workqueue(be_wq); 4141 4142 if (sriov_enabled(adapter)) 4143 be_vf_clear(adapter); 4144 4145 /* Re-configure FW to distribute resources evenly across max-supported 4146 * number of VFs, only when VFs are not already enabled. 4147 */ 4148 if (skyhawk_chip(adapter) && be_physfn(adapter) && 4149 !pci_vfs_assigned(pdev)) { 4150 be_calculate_vf_res(adapter, 4151 pci_sriov_get_totalvfs(pdev), 4152 &vft_res); 4153 be_cmd_set_sriov_config(adapter, adapter->pool_res, 4154 pci_sriov_get_totalvfs(pdev), 4155 &vft_res); 4156 } 4157 4158 be_vxlan_unset_port(adapter->netdev, 0, 0, NULL); 4159 4160 be_if_destroy(adapter); 4161 4162 be_clear_queues(adapter); 4163 4164 be_msix_disable(adapter); 4165 adapter->flags &= ~BE_FLAGS_SETUP_DONE; 4166 return 0; 4167 } 4168 4169 static int be_vfs_if_create(struct be_adapter *adapter) 4170 { 4171 struct be_resources res = {0}; 4172 u32 cap_flags, en_flags, vf; 4173 struct be_vf_cfg *vf_cfg; 4174 int status; 4175 4176 /* If a FW profile exists, then cap_flags are updated */ 4177 cap_flags = BE_VF_IF_EN_FLAGS; 4178 4179 for_all_vfs(adapter, vf_cfg, vf) { 4180 if (!BE3_chip(adapter)) { 4181 status = be_cmd_get_profile_config(adapter, &res, NULL, 4182 ACTIVE_PROFILE_TYPE, 4183 RESOURCE_LIMITS, 4184 vf + 1); 4185 if (!status) { 4186 cap_flags = res.if_cap_flags; 4187 /* Prevent VFs from enabling VLAN promiscuous 4188 * mode 4189 */ 4190 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS; 4191 } 4192 } 4193 4194 /* PF should enable IF flags during proxy if_create call */ 4195 en_flags = cap_flags & BE_VF_IF_EN_FLAGS; 4196 status = be_cmd_if_create(adapter, cap_flags, en_flags, 4197 &vf_cfg->if_handle, vf + 1); 4198 if (status) 4199 return status; 4200 } 4201 4202 return 0; 4203 } 4204 4205 static int be_vf_setup_init(struct be_adapter *adapter) 4206 { 4207 struct be_vf_cfg *vf_cfg; 4208 int vf; 4209 4210 adapter->vf_cfg = kzalloc_objs(*vf_cfg, adapter->num_vfs); 4211 if (!adapter->vf_cfg) 4212 return -ENOMEM; 4213 4214 for_all_vfs(adapter, vf_cfg, vf) { 4215 vf_cfg->if_handle = -1; 4216 vf_cfg->pmac_id = -1; 4217 } 4218 return 0; 4219 } 4220 4221 static int be_vf_setup(struct be_adapter *adapter) 4222 { 4223 struct device *dev = &adapter->pdev->dev; 4224 struct be_vf_cfg *vf_cfg; 4225 int status, old_vfs, vf; 4226 bool spoofchk; 4227 4228 old_vfs = pci_num_vf(adapter->pdev); 4229 4230 status = be_vf_setup_init(adapter); 4231 if (status) 4232 goto err; 4233 4234 if (old_vfs) { 4235 for_all_vfs(adapter, vf_cfg, vf) { 4236 status = be_cmd_get_if_id(adapter, vf_cfg, vf); 4237 if (status) 4238 goto err; 4239 } 4240 4241 status = be_vfs_mac_query(adapter); 4242 if (status) 4243 goto err; 4244 } else { 4245 status = be_vfs_if_create(adapter); 4246 if (status) 4247 goto err; 4248 4249 status = be_vf_eth_addr_config(adapter); 4250 if (status) 4251 goto err; 4252 } 4253 4254 for_all_vfs(adapter, vf_cfg, vf) { 4255 /* Allow VFs to programs MAC/VLAN filters */ 4256 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges, 4257 vf + 1); 4258 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) { 4259 status = be_cmd_set_fn_privileges(adapter, 4260 vf_cfg->privileges | 4261 BE_PRIV_FILTMGMT, 4262 vf + 1); 4263 if (!status) { 4264 vf_cfg->privileges |= BE_PRIV_FILTMGMT; 4265 dev_info(dev, "VF%d has FILTMGMT privilege\n", 4266 vf); 4267 } 4268 } 4269 4270 /* Allow full available bandwidth */ 4271 if (!old_vfs) 4272 be_cmd_config_qos(adapter, 0, 0, vf + 1); 4273 4274 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1, 4275 vf_cfg->if_handle, NULL, 4276 &spoofchk); 4277 if (!status) 4278 vf_cfg->spoofchk = spoofchk; 4279 4280 if (!old_vfs) { 4281 be_cmd_enable_vf(adapter, vf + 1); 4282 be_cmd_set_logical_link_config(adapter, 4283 IFLA_VF_LINK_STATE_AUTO, 4284 vf+1); 4285 } 4286 } 4287 4288 if (!old_vfs) { 4289 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs); 4290 if (status) { 4291 dev_err(dev, "SRIOV enable failed\n"); 4292 adapter->num_vfs = 0; 4293 goto err; 4294 } 4295 } 4296 4297 if (BE3_chip(adapter)) { 4298 /* On BE3, enable VEB only when SRIOV is enabled */ 4299 status = be_cmd_set_hsw_config(adapter, 0, 0, 4300 adapter->if_handle, 4301 PORT_FWD_TYPE_VEB, 0); 4302 if (status) 4303 goto err; 4304 } 4305 4306 adapter->flags |= BE_FLAGS_SRIOV_ENABLED; 4307 return 0; 4308 err: 4309 dev_err(dev, "VF setup failed\n"); 4310 be_vf_clear(adapter); 4311 return status; 4312 } 4313 4314 /* Converting function_mode bits on BE3 to SH mc_type enums */ 4315 4316 static u8 be_convert_mc_type(u32 function_mode) 4317 { 4318 if (function_mode & VNIC_MODE && function_mode & QNQ_MODE) 4319 return vNIC1; 4320 else if (function_mode & QNQ_MODE) 4321 return FLEX10; 4322 else if (function_mode & VNIC_MODE) 4323 return vNIC2; 4324 else if (function_mode & UMC_ENABLED) 4325 return UMC; 4326 else 4327 return MC_NONE; 4328 } 4329 4330 /* On BE2/BE3 FW does not suggest the supported limits */ 4331 static void BEx_get_resources(struct be_adapter *adapter, 4332 struct be_resources *res) 4333 { 4334 bool use_sriov = adapter->num_vfs ? 1 : 0; 4335 4336 if (be_physfn(adapter)) 4337 res->max_uc_mac = BE_UC_PMAC_COUNT; 4338 else 4339 res->max_uc_mac = BE_VF_UC_PMAC_COUNT; 4340 4341 adapter->mc_type = be_convert_mc_type(adapter->function_mode); 4342 4343 if (be_is_mc(adapter)) { 4344 /* Assuming that there are 4 channels per port, 4345 * when multi-channel is enabled 4346 */ 4347 if (be_is_qnq_mode(adapter)) 4348 res->max_vlans = BE_NUM_VLANS_SUPPORTED/8; 4349 else 4350 /* In a non-qnq multichannel mode, the pvid 4351 * takes up one vlan entry 4352 */ 4353 res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1; 4354 } else { 4355 res->max_vlans = BE_NUM_VLANS_SUPPORTED; 4356 } 4357 4358 res->max_mcast_mac = BE_MAX_MC; 4359 4360 /* 1) For BE3 1Gb ports, FW does not support multiple TXQs 4361 * 2) Create multiple TX rings on a BE3-R multi-channel interface 4362 * *only* if it is RSS-capable. 4363 */ 4364 if (BE2_chip(adapter) || use_sriov || (adapter->port_num > 1) || 4365 be_virtfn(adapter) || 4366 (be_is_mc(adapter) && 4367 !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) { 4368 res->max_tx_qs = 1; 4369 } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) { 4370 struct be_resources super_nic_res = {0}; 4371 4372 /* On a SuperNIC profile, the driver needs to use the 4373 * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits 4374 */ 4375 be_cmd_get_profile_config(adapter, &super_nic_res, NULL, 4376 ACTIVE_PROFILE_TYPE, RESOURCE_LIMITS, 4377 0); 4378 /* Some old versions of BE3 FW don't report max_tx_qs value */ 4379 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS; 4380 } else { 4381 res->max_tx_qs = BE3_MAX_TX_QS; 4382 } 4383 4384 if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) && 4385 !use_sriov && be_physfn(adapter)) 4386 res->max_rss_qs = (adapter->be3_native) ? 4387 BE3_MAX_RSS_QS : BE2_MAX_RSS_QS; 4388 res->max_rx_qs = res->max_rss_qs + 1; 4389 4390 if (be_physfn(adapter)) 4391 res->max_evt_qs = (be_max_vfs(adapter) > 0) ? 4392 BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS; 4393 else 4394 res->max_evt_qs = 1; 4395 4396 res->if_cap_flags = BE_IF_CAP_FLAGS_WANT; 4397 res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS; 4398 if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS)) 4399 res->if_cap_flags &= ~BE_IF_FLAGS_RSS; 4400 } 4401 4402 static void be_setup_init(struct be_adapter *adapter) 4403 { 4404 adapter->vlan_prio_bmap = 0xff; 4405 adapter->phy.link_speed = -1; 4406 adapter->if_handle = -1; 4407 adapter->be3_native = false; 4408 adapter->if_flags = 0; 4409 adapter->phy_state = BE_UNKNOWN_PHY_STATE; 4410 if (be_physfn(adapter)) 4411 adapter->cmd_privileges = MAX_PRIVILEGES; 4412 else 4413 adapter->cmd_privileges = MIN_PRIVILEGES; 4414 } 4415 4416 /* HW supports only MAX_PORT_RSS_TABLES RSS Policy Tables per port. 4417 * However, this HW limitation is not exposed to the host via any SLI cmd. 4418 * As a result, in the case of SRIOV and in particular multi-partition configs 4419 * the driver needs to calculate a proportional share of RSS Tables per PF-pool 4420 * for distribution between the VFs. This self-imposed limit will determine the 4421 * no: of VFs for which RSS can be enabled. 4422 */ 4423 static void be_calculate_pf_pool_rss_tables(struct be_adapter *adapter) 4424 { 4425 struct be_port_resources port_res = {0}; 4426 u8 rss_tables_on_port; 4427 u16 max_vfs = be_max_vfs(adapter); 4428 4429 be_cmd_get_profile_config(adapter, NULL, &port_res, SAVED_PROFILE_TYPE, 4430 RESOURCE_LIMITS, 0); 4431 4432 rss_tables_on_port = MAX_PORT_RSS_TABLES - port_res.nic_pfs; 4433 4434 /* Each PF Pool's RSS Tables limit = 4435 * PF's Max VFs / Total_Max_VFs on Port * RSS Tables on Port 4436 */ 4437 adapter->pool_res.max_rss_tables = 4438 max_vfs * rss_tables_on_port / port_res.max_vfs; 4439 } 4440 4441 static int be_get_sriov_config(struct be_adapter *adapter) 4442 { 4443 struct be_resources res = {0}; 4444 int max_vfs, old_vfs; 4445 4446 be_cmd_get_profile_config(adapter, &res, NULL, ACTIVE_PROFILE_TYPE, 4447 RESOURCE_LIMITS, 0); 4448 4449 /* Some old versions of BE3 FW don't report max_vfs value */ 4450 if (BE3_chip(adapter) && !res.max_vfs) { 4451 max_vfs = pci_sriov_get_totalvfs(adapter->pdev); 4452 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0; 4453 } 4454 4455 adapter->pool_res = res; 4456 4457 /* If during previous unload of the driver, the VFs were not disabled, 4458 * then we cannot rely on the PF POOL limits for the TotalVFs value. 4459 * Instead use the TotalVFs value stored in the pci-dev struct. 4460 */ 4461 old_vfs = pci_num_vf(adapter->pdev); 4462 if (old_vfs) { 4463 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n", 4464 old_vfs); 4465 4466 adapter->pool_res.max_vfs = 4467 pci_sriov_get_totalvfs(adapter->pdev); 4468 adapter->num_vfs = old_vfs; 4469 } 4470 4471 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) { 4472 be_calculate_pf_pool_rss_tables(adapter); 4473 dev_info(&adapter->pdev->dev, 4474 "RSS can be enabled for all VFs if num_vfs <= %d\n", 4475 be_max_pf_pool_rss_tables(adapter)); 4476 } 4477 return 0; 4478 } 4479 4480 static void be_alloc_sriov_res(struct be_adapter *adapter) 4481 { 4482 int old_vfs = pci_num_vf(adapter->pdev); 4483 struct be_resources vft_res = {0}; 4484 int status; 4485 4486 be_get_sriov_config(adapter); 4487 4488 if (!old_vfs) 4489 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter)); 4490 4491 /* When the HW is in SRIOV capable configuration, the PF-pool 4492 * resources are given to PF during driver load, if there are no 4493 * old VFs. This facility is not available in BE3 FW. 4494 * Also, this is done by FW in Lancer chip. 4495 */ 4496 if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) { 4497 be_calculate_vf_res(adapter, 0, &vft_res); 4498 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0, 4499 &vft_res); 4500 if (status) 4501 dev_err(&adapter->pdev->dev, 4502 "Failed to optimize SRIOV resources\n"); 4503 } 4504 } 4505 4506 static int be_get_resources(struct be_adapter *adapter) 4507 { 4508 struct device *dev = &adapter->pdev->dev; 4509 struct be_resources res = {0}; 4510 int status; 4511 4512 /* For Lancer, SH etc read per-function resource limits from FW. 4513 * GET_FUNC_CONFIG returns per function guaranteed limits. 4514 * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits 4515 */ 4516 if (BEx_chip(adapter)) { 4517 BEx_get_resources(adapter, &res); 4518 } else { 4519 status = be_cmd_get_func_config(adapter, &res); 4520 if (status) 4521 return status; 4522 4523 /* If a default RXQ must be created, we'll use up one RSSQ*/ 4524 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs && 4525 !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS)) 4526 res.max_rss_qs -= 1; 4527 } 4528 4529 /* If RoCE is supported stash away half the EQs for RoCE */ 4530 res.max_nic_evt_qs = be_roce_supported(adapter) ? 4531 res.max_evt_qs / 2 : res.max_evt_qs; 4532 adapter->res = res; 4533 4534 /* If FW supports RSS default queue, then skip creating non-RSS 4535 * queue for non-IP traffic. 4536 */ 4537 adapter->need_def_rxq = (be_if_cap_flags(adapter) & 4538 BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1; 4539 4540 dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n", 4541 be_max_txqs(adapter), be_max_rxqs(adapter), 4542 be_max_rss(adapter), be_max_nic_eqs(adapter), 4543 be_max_vfs(adapter)); 4544 dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n", 4545 be_max_uc(adapter), be_max_mc(adapter), 4546 be_max_vlans(adapter)); 4547 4548 /* Ensure RX and TX queues are created in pairs at init time */ 4549 adapter->cfg_num_rx_irqs = 4550 min_t(u16, netif_get_num_default_rss_queues(), 4551 be_max_qp_irqs(adapter)); 4552 adapter->cfg_num_tx_irqs = adapter->cfg_num_rx_irqs; 4553 return 0; 4554 } 4555 4556 static int be_get_config(struct be_adapter *adapter) 4557 { 4558 int status, level; 4559 u16 profile_id; 4560 4561 status = be_cmd_get_cntl_attributes(adapter); 4562 if (status) 4563 return status; 4564 4565 status = be_cmd_query_fw_cfg(adapter); 4566 if (status) 4567 return status; 4568 4569 if (!lancer_chip(adapter) && be_physfn(adapter)) 4570 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len); 4571 4572 if (BEx_chip(adapter)) { 4573 level = be_cmd_get_fw_log_level(adapter); 4574 adapter->msg_enable = 4575 level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0; 4576 } 4577 4578 be_cmd_get_acpi_wol_cap(adapter); 4579 pci_enable_wake(adapter->pdev, PCI_D3hot, adapter->wol_en); 4580 pci_enable_wake(adapter->pdev, PCI_D3cold, adapter->wol_en); 4581 4582 be_cmd_query_port_name(adapter); 4583 4584 if (be_physfn(adapter)) { 4585 status = be_cmd_get_active_profile(adapter, &profile_id); 4586 if (!status) 4587 dev_info(&adapter->pdev->dev, 4588 "Using profile 0x%x\n", profile_id); 4589 } 4590 4591 return 0; 4592 } 4593 4594 static int be_mac_setup(struct be_adapter *adapter) 4595 { 4596 u8 mac[ETH_ALEN]; 4597 int status; 4598 4599 if (is_zero_ether_addr(adapter->netdev->dev_addr)) { 4600 status = be_cmd_get_perm_mac(adapter, mac); 4601 if (status) 4602 return status; 4603 4604 eth_hw_addr_set(adapter->netdev, mac); 4605 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN); 4606 4607 /* Initial MAC for BE3 VFs is already programmed by PF */ 4608 if (BEx_chip(adapter) && be_virtfn(adapter)) 4609 memcpy(adapter->dev_mac, mac, ETH_ALEN); 4610 } 4611 4612 return 0; 4613 } 4614 4615 static void be_schedule_worker(struct be_adapter *adapter) 4616 { 4617 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000)); 4618 adapter->flags |= BE_FLAGS_WORKER_SCHEDULED; 4619 } 4620 4621 static void be_destroy_err_recovery_workq(void) 4622 { 4623 if (!be_err_recovery_workq) 4624 return; 4625 4626 destroy_workqueue(be_err_recovery_workq); 4627 be_err_recovery_workq = NULL; 4628 } 4629 4630 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay) 4631 { 4632 struct be_error_recovery *err_rec = &adapter->error_recovery; 4633 4634 if (!be_err_recovery_workq) 4635 return; 4636 4637 queue_delayed_work(be_err_recovery_workq, &err_rec->err_detection_work, 4638 msecs_to_jiffies(delay)); 4639 adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED; 4640 } 4641 4642 static int be_setup_queues(struct be_adapter *adapter) 4643 { 4644 struct net_device *netdev = adapter->netdev; 4645 int status; 4646 4647 status = be_evt_queues_create(adapter); 4648 if (status) 4649 goto err; 4650 4651 status = be_tx_qs_create(adapter); 4652 if (status) 4653 goto err; 4654 4655 status = be_rx_cqs_create(adapter); 4656 if (status) 4657 goto err; 4658 4659 status = be_mcc_queues_create(adapter); 4660 if (status) 4661 goto err; 4662 4663 status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs); 4664 if (status) 4665 goto err; 4666 4667 status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs); 4668 if (status) 4669 goto err; 4670 4671 return 0; 4672 err: 4673 dev_err(&adapter->pdev->dev, "queue_setup failed\n"); 4674 return status; 4675 } 4676 4677 static int be_if_create(struct be_adapter *adapter) 4678 { 4679 u32 en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS; 4680 u32 cap_flags = be_if_cap_flags(adapter); 4681 4682 /* alloc required memory for other filtering fields */ 4683 adapter->pmac_id = kcalloc(be_max_uc(adapter), 4684 sizeof(*adapter->pmac_id), GFP_KERNEL); 4685 if (!adapter->pmac_id) 4686 return -ENOMEM; 4687 4688 adapter->mc_list = kzalloc_objs(*adapter->mc_list, be_max_mc(adapter)); 4689 if (!adapter->mc_list) 4690 return -ENOMEM; 4691 4692 adapter->uc_list = kzalloc_objs(*adapter->uc_list, be_max_uc(adapter)); 4693 if (!adapter->uc_list) 4694 return -ENOMEM; 4695 4696 if (adapter->cfg_num_rx_irqs == 1) 4697 cap_flags &= ~(BE_IF_FLAGS_DEFQ_RSS | BE_IF_FLAGS_RSS); 4698 4699 en_flags &= cap_flags; 4700 /* will enable all the needed filter flags in be_open() */ 4701 return be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags, 4702 &adapter->if_handle, 0); 4703 } 4704 4705 int be_update_queues(struct be_adapter *adapter) 4706 { 4707 struct net_device *netdev = adapter->netdev; 4708 int status; 4709 4710 if (netif_running(netdev)) { 4711 /* be_tx_timeout() must not run concurrently with this 4712 * function, synchronize with an already-running dev_watchdog 4713 */ 4714 netif_tx_lock_bh(netdev); 4715 /* device cannot transmit now, avoid dev_watchdog timeouts */ 4716 netif_carrier_off(netdev); 4717 netif_tx_unlock_bh(netdev); 4718 4719 be_close(netdev); 4720 } 4721 4722 be_cancel_worker(adapter); 4723 4724 /* If any vectors have been shared with RoCE we cannot re-program 4725 * the MSIx table. 4726 */ 4727 if (!adapter->num_msix_roce_vec) 4728 be_msix_disable(adapter); 4729 4730 be_clear_queues(adapter); 4731 status = be_cmd_if_destroy(adapter, adapter->if_handle, 0); 4732 if (status) 4733 return status; 4734 4735 if (!msix_enabled(adapter)) { 4736 status = be_msix_enable(adapter); 4737 if (status) 4738 return status; 4739 } 4740 4741 status = be_if_create(adapter); 4742 if (status) 4743 return status; 4744 4745 status = be_setup_queues(adapter); 4746 if (status) 4747 return status; 4748 4749 be_schedule_worker(adapter); 4750 4751 /* The IF was destroyed and re-created. We need to clear 4752 * all promiscuous flags valid for the destroyed IF. 4753 * Without this promisc mode is not restored during 4754 * be_open() because the driver thinks that it is 4755 * already enabled in HW. 4756 */ 4757 adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS; 4758 4759 if (netif_running(netdev)) 4760 status = be_open(netdev); 4761 4762 return status; 4763 } 4764 4765 static inline int fw_major_num(const char *fw_ver) 4766 { 4767 int fw_major = 0, i; 4768 4769 i = sscanf(fw_ver, "%d.", &fw_major); 4770 if (i != 1) 4771 return 0; 4772 4773 return fw_major; 4774 } 4775 4776 /* If it is error recovery, FLR the PF 4777 * Else if any VFs are already enabled don't FLR the PF 4778 */ 4779 static bool be_reset_required(struct be_adapter *adapter) 4780 { 4781 if (be_error_recovering(adapter)) 4782 return true; 4783 else 4784 return pci_num_vf(adapter->pdev) == 0; 4785 } 4786 4787 /* Wait for the FW to be ready and perform the required initialization */ 4788 static int be_func_init(struct be_adapter *adapter) 4789 { 4790 int status; 4791 4792 status = be_fw_wait_ready(adapter); 4793 if (status) 4794 return status; 4795 4796 /* FW is now ready; clear errors to allow cmds/doorbell */ 4797 be_clear_error(adapter, BE_CLEAR_ALL); 4798 4799 if (be_reset_required(adapter)) { 4800 status = be_cmd_reset_function(adapter); 4801 if (status) 4802 return status; 4803 4804 /* Wait for interrupts to quiesce after an FLR */ 4805 msleep(100); 4806 } 4807 4808 /* Tell FW we're ready to fire cmds */ 4809 status = be_cmd_fw_init(adapter); 4810 if (status) 4811 return status; 4812 4813 /* Allow interrupts for other ULPs running on NIC function */ 4814 be_intr_set(adapter, true); 4815 4816 return 0; 4817 } 4818 4819 static int be_setup(struct be_adapter *adapter) 4820 { 4821 struct device *dev = &adapter->pdev->dev; 4822 int status; 4823 4824 status = be_func_init(adapter); 4825 if (status) 4826 return status; 4827 4828 be_setup_init(adapter); 4829 4830 if (!lancer_chip(adapter)) 4831 be_cmd_req_native_mode(adapter); 4832 4833 /* invoke this cmd first to get pf_num and vf_num which are needed 4834 * for issuing profile related cmds 4835 */ 4836 if (!BEx_chip(adapter)) { 4837 status = be_cmd_get_func_config(adapter, NULL); 4838 if (status) 4839 return status; 4840 } 4841 4842 status = be_get_config(adapter); 4843 if (status) 4844 goto err; 4845 4846 if (!BE2_chip(adapter) && be_physfn(adapter)) 4847 be_alloc_sriov_res(adapter); 4848 4849 status = be_get_resources(adapter); 4850 if (status) 4851 goto err; 4852 4853 status = be_msix_enable(adapter); 4854 if (status) 4855 goto err; 4856 4857 /* will enable all the needed filter flags in be_open() */ 4858 status = be_if_create(adapter); 4859 if (status) 4860 goto err; 4861 4862 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */ 4863 rtnl_lock(); 4864 status = be_setup_queues(adapter); 4865 rtnl_unlock(); 4866 if (status) 4867 goto err; 4868 4869 be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0); 4870 4871 status = be_mac_setup(adapter); 4872 if (status) 4873 goto err; 4874 4875 be_cmd_get_fw_ver(adapter); 4876 dev_info(dev, "FW version is %s\n", adapter->fw_ver); 4877 4878 if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) { 4879 dev_err(dev, "Firmware on card is old(%s), IRQs may not work", 4880 adapter->fw_ver); 4881 dev_err(dev, "Please upgrade firmware to version >= 4.0\n"); 4882 } 4883 4884 status = be_cmd_set_flow_control(adapter, adapter->tx_fc, 4885 adapter->rx_fc); 4886 if (status) 4887 be_cmd_get_flow_control(adapter, &adapter->tx_fc, 4888 &adapter->rx_fc); 4889 4890 dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n", 4891 adapter->tx_fc, adapter->rx_fc); 4892 4893 if (be_physfn(adapter)) 4894 be_cmd_set_logical_link_config(adapter, 4895 IFLA_VF_LINK_STATE_AUTO, 0); 4896 4897 /* BE3 EVB echoes broadcast/multicast packets back to PF's vport 4898 * confusing a linux bridge or OVS that it might be connected to. 4899 * Set the EVB to PASSTHRU mode which effectively disables the EVB 4900 * when SRIOV is not enabled. 4901 */ 4902 if (BE3_chip(adapter)) 4903 be_cmd_set_hsw_config(adapter, 0, 0, adapter->if_handle, 4904 PORT_FWD_TYPE_PASSTHRU, 0); 4905 4906 if (adapter->num_vfs) 4907 be_vf_setup(adapter); 4908 4909 status = be_cmd_get_phy_info(adapter); 4910 if (!status && be_pause_supported(adapter)) 4911 adapter->phy.fc_autoneg = 1; 4912 4913 if (be_physfn(adapter) && !lancer_chip(adapter)) 4914 be_cmd_set_features(adapter); 4915 4916 be_schedule_worker(adapter); 4917 adapter->flags |= BE_FLAGS_SETUP_DONE; 4918 return 0; 4919 err: 4920 be_clear(adapter); 4921 return status; 4922 } 4923 4924 #ifdef CONFIG_NET_POLL_CONTROLLER 4925 static void be_netpoll(struct net_device *netdev) 4926 { 4927 struct be_adapter *adapter = netdev_priv(netdev); 4928 struct be_eq_obj *eqo; 4929 int i; 4930 4931 for_all_evt_queues(adapter, eqo, i) { 4932 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0); 4933 napi_schedule(&eqo->napi); 4934 } 4935 } 4936 #endif 4937 4938 int be_load_fw(struct be_adapter *adapter, u8 *fw_file) 4939 { 4940 const struct firmware *fw; 4941 int status; 4942 4943 if (!netif_running(adapter->netdev)) { 4944 dev_err(&adapter->pdev->dev, 4945 "Firmware load not allowed (interface is down)\n"); 4946 return -ENETDOWN; 4947 } 4948 4949 status = request_firmware(&fw, fw_file, &adapter->pdev->dev); 4950 if (status) 4951 goto fw_exit; 4952 4953 dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file); 4954 4955 if (lancer_chip(adapter)) 4956 status = lancer_fw_download(adapter, fw); 4957 else 4958 status = be_fw_download(adapter, fw); 4959 4960 if (!status) 4961 be_cmd_get_fw_ver(adapter); 4962 4963 fw_exit: 4964 release_firmware(fw); 4965 return status; 4966 } 4967 4968 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh, 4969 u16 flags, struct netlink_ext_ack *extack) 4970 { 4971 struct be_adapter *adapter = netdev_priv(dev); 4972 struct nlattr *attr, *br_spec; 4973 int rem; 4974 int status = 0; 4975 u16 mode = 0; 4976 4977 if (!sriov_enabled(adapter)) 4978 return -EOPNOTSUPP; 4979 4980 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 4981 if (!br_spec) 4982 return -EINVAL; 4983 4984 nla_for_each_nested_type(attr, IFLA_BRIDGE_MODE, br_spec, rem) { 4985 mode = nla_get_u16(attr); 4986 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA) 4987 return -EOPNOTSUPP; 4988 4989 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB) 4990 return -EINVAL; 4991 4992 status = be_cmd_set_hsw_config(adapter, 0, 0, 4993 adapter->if_handle, 4994 mode == BRIDGE_MODE_VEPA ? 4995 PORT_FWD_TYPE_VEPA : 4996 PORT_FWD_TYPE_VEB, 0); 4997 if (status) 4998 goto err; 4999 5000 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n", 5001 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB"); 5002 5003 return status; 5004 } 5005 err: 5006 dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n", 5007 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB"); 5008 5009 return status; 5010 } 5011 5012 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 5013 struct net_device *dev, u32 filter_mask, 5014 int nlflags) 5015 { 5016 struct be_adapter *adapter = netdev_priv(dev); 5017 int status = 0; 5018 u8 hsw_mode; 5019 5020 /* BE and Lancer chips support VEB mode only */ 5021 if (BEx_chip(adapter) || lancer_chip(adapter)) { 5022 /* VEB is disabled in non-SR-IOV profiles on BE3/Lancer */ 5023 if (!pci_sriov_get_totalvfs(adapter->pdev)) 5024 return 0; 5025 hsw_mode = PORT_FWD_TYPE_VEB; 5026 } else { 5027 status = be_cmd_get_hsw_config(adapter, NULL, 0, 5028 adapter->if_handle, &hsw_mode, 5029 NULL); 5030 if (status) 5031 return 0; 5032 5033 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU) 5034 return 0; 5035 } 5036 5037 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, 5038 hsw_mode == PORT_FWD_TYPE_VEPA ? 5039 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB, 5040 0, 0, nlflags, filter_mask, NULL); 5041 } 5042 5043 static struct be_cmd_work *be_alloc_work(struct be_adapter *adapter, 5044 void (*func)(struct work_struct *)) 5045 { 5046 struct be_cmd_work *work; 5047 5048 work = kzalloc_obj(*work, GFP_ATOMIC); 5049 if (!work) { 5050 dev_err(&adapter->pdev->dev, 5051 "be_work memory allocation failed\n"); 5052 return NULL; 5053 } 5054 5055 INIT_WORK(&work->work, func); 5056 work->adapter = adapter; 5057 return work; 5058 } 5059 5060 static netdev_features_t be_features_check(struct sk_buff *skb, 5061 struct net_device *dev, 5062 netdev_features_t features) 5063 { 5064 struct be_adapter *adapter = netdev_priv(dev); 5065 u8 l4_hdr = 0; 5066 5067 if (skb_is_gso(skb)) { 5068 /* IPv6 TSO requests with extension hdrs are a problem 5069 * to Lancer and BE3 HW. Disable TSO6 feature. 5070 */ 5071 if (!skyhawk_chip(adapter) && is_ipv6_ext_hdr(skb)) 5072 features &= ~NETIF_F_TSO6; 5073 5074 /* Lancer cannot handle the packet with MSS less than 256. 5075 * Also it can't handle a TSO packet with a single segment 5076 * Disable the GSO support in such cases 5077 */ 5078 if (lancer_chip(adapter) && 5079 (skb_shinfo(skb)->gso_size < 256 || 5080 skb_shinfo(skb)->gso_segs == 1)) 5081 features &= ~NETIF_F_GSO_MASK; 5082 } 5083 5084 /* The code below restricts offload features for some tunneled and 5085 * Q-in-Q packets. 5086 * Offload features for normal (non tunnel) packets are unchanged. 5087 */ 5088 features = vlan_features_check(skb, features); 5089 if (!skb->encapsulation || 5090 !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)) 5091 return features; 5092 5093 /* It's an encapsulated packet and VxLAN offloads are enabled. We 5094 * should disable tunnel offload features if it's not a VxLAN packet, 5095 * as tunnel offloads have been enabled only for VxLAN. This is done to 5096 * allow other tunneled traffic like GRE work fine while VxLAN 5097 * offloads are configured in Skyhawk-R. 5098 */ 5099 switch (vlan_get_protocol(skb)) { 5100 case htons(ETH_P_IP): 5101 l4_hdr = ip_hdr(skb)->protocol; 5102 break; 5103 case htons(ETH_P_IPV6): 5104 l4_hdr = ipv6_hdr(skb)->nexthdr; 5105 break; 5106 default: 5107 return features; 5108 } 5109 5110 if (l4_hdr != IPPROTO_UDP || 5111 skb->inner_protocol_type != ENCAP_TYPE_ETHER || 5112 skb->inner_protocol != htons(ETH_P_TEB) || 5113 skb_inner_mac_header(skb) - skb_transport_header(skb) != 5114 sizeof(struct udphdr) + sizeof(struct vxlanhdr) || 5115 !adapter->vxlan_port || 5116 udp_hdr(skb)->dest != adapter->vxlan_port) 5117 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 5118 5119 return features; 5120 } 5121 5122 static int be_get_phys_port_id(struct net_device *dev, 5123 struct netdev_phys_item_id *ppid) 5124 { 5125 int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1; 5126 struct be_adapter *adapter = netdev_priv(dev); 5127 u8 *id; 5128 5129 if (MAX_PHYS_ITEM_ID_LEN < id_len) 5130 return -ENOSPC; 5131 5132 ppid->id[0] = adapter->hba_port_num + 1; 5133 id = &ppid->id[1]; 5134 for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0; 5135 i--, id += CNTL_SERIAL_NUM_WORD_SZ) 5136 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ); 5137 5138 ppid->id_len = id_len; 5139 5140 return 0; 5141 } 5142 5143 static void be_set_rx_mode(struct net_device *dev) 5144 { 5145 struct be_adapter *adapter = netdev_priv(dev); 5146 struct be_cmd_work *work; 5147 5148 work = be_alloc_work(adapter, be_work_set_rx_mode); 5149 if (work) 5150 queue_work(be_wq, &work->work); 5151 } 5152 5153 static const struct net_device_ops be_netdev_ops = { 5154 .ndo_open = be_open, 5155 .ndo_stop = be_close, 5156 .ndo_start_xmit = be_xmit, 5157 .ndo_set_rx_mode = be_set_rx_mode, 5158 .ndo_set_mac_address = be_mac_addr_set, 5159 .ndo_get_stats64 = be_get_stats64, 5160 .ndo_validate_addr = eth_validate_addr, 5161 .ndo_vlan_rx_add_vid = be_vlan_add_vid, 5162 .ndo_vlan_rx_kill_vid = be_vlan_rem_vid, 5163 .ndo_set_vf_mac = be_set_vf_mac, 5164 .ndo_set_vf_vlan = be_set_vf_vlan, 5165 .ndo_set_vf_rate = be_set_vf_tx_rate, 5166 .ndo_get_vf_config = be_get_vf_config, 5167 .ndo_set_vf_link_state = be_set_vf_link_state, 5168 .ndo_set_vf_spoofchk = be_set_vf_spoofchk, 5169 .ndo_tx_timeout = be_tx_timeout, 5170 #ifdef CONFIG_NET_POLL_CONTROLLER 5171 .ndo_poll_controller = be_netpoll, 5172 #endif 5173 .ndo_bridge_setlink = be_ndo_bridge_setlink, 5174 .ndo_bridge_getlink = be_ndo_bridge_getlink, 5175 .ndo_features_check = be_features_check, 5176 .ndo_get_phys_port_id = be_get_phys_port_id, 5177 }; 5178 5179 static void be_netdev_init(struct net_device *netdev) 5180 { 5181 struct be_adapter *adapter = netdev_priv(netdev); 5182 5183 netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 | 5184 NETIF_F_GSO_UDP_TUNNEL | 5185 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM | 5186 NETIF_F_HW_VLAN_CTAG_TX; 5187 if ((be_if_cap_flags(adapter) & BE_IF_FLAGS_RSS)) 5188 netdev->hw_features |= NETIF_F_RXHASH; 5189 5190 netdev->features |= netdev->hw_features | 5191 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER | 5192 NETIF_F_HIGHDMA; 5193 5194 netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 | 5195 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 5196 5197 netdev->priv_flags |= IFF_UNICAST_FLT; 5198 5199 netdev->flags |= IFF_MULTICAST; 5200 5201 netif_set_tso_max_size(netdev, BE_MAX_GSO_SIZE - ETH_HLEN); 5202 5203 netdev->netdev_ops = &be_netdev_ops; 5204 5205 netdev->ethtool_ops = &be_ethtool_ops; 5206 5207 if (!lancer_chip(adapter) && !BEx_chip(adapter) && !be_is_mc(adapter)) 5208 netdev->udp_tunnel_nic_info = &be_udp_tunnels; 5209 5210 /* MTU range: 256 - 9000 */ 5211 netdev->min_mtu = BE_MIN_MTU; 5212 netdev->max_mtu = BE_MAX_MTU; 5213 } 5214 5215 static void be_cleanup(struct be_adapter *adapter) 5216 { 5217 struct net_device *netdev = adapter->netdev; 5218 5219 rtnl_lock(); 5220 netif_device_detach(netdev); 5221 if (netif_running(netdev)) 5222 be_close(netdev); 5223 rtnl_unlock(); 5224 5225 be_clear(adapter); 5226 } 5227 5228 static int be_resume(struct be_adapter *adapter) 5229 { 5230 struct net_device *netdev = adapter->netdev; 5231 int status; 5232 5233 status = be_setup(adapter); 5234 if (status) 5235 return status; 5236 5237 rtnl_lock(); 5238 if (netif_running(netdev)) 5239 status = be_open(netdev); 5240 rtnl_unlock(); 5241 5242 if (status) 5243 return status; 5244 5245 netif_device_attach(netdev); 5246 5247 return 0; 5248 } 5249 5250 static void be_soft_reset(struct be_adapter *adapter) 5251 { 5252 u32 val; 5253 5254 dev_info(&adapter->pdev->dev, "Initiating chip soft reset\n"); 5255 val = ioread32(adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET); 5256 val |= SLIPORT_SOFTRESET_SR_MASK; 5257 iowrite32(val, adapter->pcicfg + SLIPORT_SOFTRESET_OFFSET); 5258 } 5259 5260 static bool be_err_is_recoverable(struct be_adapter *adapter) 5261 { 5262 struct be_error_recovery *err_rec = &adapter->error_recovery; 5263 unsigned long initial_idle_time = 5264 msecs_to_jiffies(ERR_RECOVERY_IDLE_TIME); 5265 unsigned long recovery_interval = 5266 msecs_to_jiffies(ERR_RECOVERY_INTERVAL); 5267 u16 ue_err_code; 5268 u32 val; 5269 5270 val = be_POST_stage_get(adapter); 5271 if ((val & POST_STAGE_RECOVERABLE_ERR) != POST_STAGE_RECOVERABLE_ERR) 5272 return false; 5273 ue_err_code = val & POST_ERR_RECOVERY_CODE_MASK; 5274 if (ue_err_code == 0) 5275 return false; 5276 5277 dev_err(&adapter->pdev->dev, "Recoverable HW error code: 0x%x\n", 5278 ue_err_code); 5279 5280 if (time_before_eq(jiffies - err_rec->probe_time, initial_idle_time)) { 5281 dev_err(&adapter->pdev->dev, 5282 "Cannot recover within %lu sec from driver load\n", 5283 jiffies_to_msecs(initial_idle_time) / MSEC_PER_SEC); 5284 return false; 5285 } 5286 5287 if (err_rec->last_recovery_time && time_before_eq( 5288 jiffies - err_rec->last_recovery_time, recovery_interval)) { 5289 dev_err(&adapter->pdev->dev, 5290 "Cannot recover within %lu sec from last recovery\n", 5291 jiffies_to_msecs(recovery_interval) / MSEC_PER_SEC); 5292 return false; 5293 } 5294 5295 if (ue_err_code == err_rec->last_err_code) { 5296 dev_err(&adapter->pdev->dev, 5297 "Cannot recover from a consecutive TPE error\n"); 5298 return false; 5299 } 5300 5301 err_rec->last_recovery_time = jiffies; 5302 err_rec->last_err_code = ue_err_code; 5303 return true; 5304 } 5305 5306 static int be_tpe_recover(struct be_adapter *adapter) 5307 { 5308 struct be_error_recovery *err_rec = &adapter->error_recovery; 5309 int status = -EAGAIN; 5310 u32 val; 5311 5312 switch (err_rec->recovery_state) { 5313 case ERR_RECOVERY_ST_NONE: 5314 err_rec->recovery_state = ERR_RECOVERY_ST_DETECT; 5315 err_rec->resched_delay = ERR_RECOVERY_UE_DETECT_DURATION; 5316 break; 5317 5318 case ERR_RECOVERY_ST_DETECT: 5319 val = be_POST_stage_get(adapter); 5320 if ((val & POST_STAGE_RECOVERABLE_ERR) != 5321 POST_STAGE_RECOVERABLE_ERR) { 5322 dev_err(&adapter->pdev->dev, 5323 "Unrecoverable HW error detected: 0x%x\n", val); 5324 status = -EINVAL; 5325 err_rec->resched_delay = 0; 5326 break; 5327 } 5328 5329 dev_err(&adapter->pdev->dev, "Recoverable HW error detected\n"); 5330 5331 /* Only PF0 initiates Chip Soft Reset. But PF0 must wait UE2SR 5332 * milliseconds before it checks for final error status in 5333 * SLIPORT_SEMAPHORE to determine if recovery criteria is met. 5334 * If it does, then PF0 initiates a Soft Reset. 5335 */ 5336 if (adapter->pf_num == 0) { 5337 err_rec->recovery_state = ERR_RECOVERY_ST_RESET; 5338 err_rec->resched_delay = err_rec->ue_to_reset_time - 5339 ERR_RECOVERY_UE_DETECT_DURATION; 5340 break; 5341 } 5342 5343 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL; 5344 err_rec->resched_delay = err_rec->ue_to_poll_time - 5345 ERR_RECOVERY_UE_DETECT_DURATION; 5346 break; 5347 5348 case ERR_RECOVERY_ST_RESET: 5349 if (!be_err_is_recoverable(adapter)) { 5350 dev_err(&adapter->pdev->dev, 5351 "Failed to meet recovery criteria\n"); 5352 status = -EIO; 5353 err_rec->resched_delay = 0; 5354 break; 5355 } 5356 be_soft_reset(adapter); 5357 err_rec->recovery_state = ERR_RECOVERY_ST_PRE_POLL; 5358 err_rec->resched_delay = err_rec->ue_to_poll_time - 5359 err_rec->ue_to_reset_time; 5360 break; 5361 5362 case ERR_RECOVERY_ST_PRE_POLL: 5363 err_rec->recovery_state = ERR_RECOVERY_ST_REINIT; 5364 err_rec->resched_delay = 0; 5365 status = 0; /* done */ 5366 break; 5367 5368 default: 5369 status = -EINVAL; 5370 err_rec->resched_delay = 0; 5371 break; 5372 } 5373 5374 return status; 5375 } 5376 5377 static int be_err_recover(struct be_adapter *adapter) 5378 { 5379 int status; 5380 5381 if (!lancer_chip(adapter)) { 5382 if (!adapter->error_recovery.recovery_supported || 5383 adapter->priv_flags & BE_DISABLE_TPE_RECOVERY) 5384 return -EIO; 5385 status = be_tpe_recover(adapter); 5386 if (status) 5387 goto err; 5388 } 5389 5390 /* Wait for adapter to reach quiescent state before 5391 * destroying queues 5392 */ 5393 status = be_fw_wait_ready(adapter); 5394 if (status) 5395 goto err; 5396 5397 adapter->flags |= BE_FLAGS_TRY_RECOVERY; 5398 5399 be_cleanup(adapter); 5400 5401 status = be_resume(adapter); 5402 if (status) 5403 goto err; 5404 5405 adapter->flags &= ~BE_FLAGS_TRY_RECOVERY; 5406 5407 err: 5408 return status; 5409 } 5410 5411 static void be_err_detection_task(struct work_struct *work) 5412 { 5413 struct be_error_recovery *err_rec = 5414 container_of(work, struct be_error_recovery, 5415 err_detection_work.work); 5416 struct be_adapter *adapter = 5417 container_of(err_rec, struct be_adapter, 5418 error_recovery); 5419 u32 resched_delay = ERR_RECOVERY_DETECTION_DELAY; 5420 struct device *dev = &adapter->pdev->dev; 5421 int recovery_status; 5422 5423 be_detect_error(adapter); 5424 if (!be_check_error(adapter, BE_ERROR_HW)) 5425 goto reschedule_task; 5426 5427 recovery_status = be_err_recover(adapter); 5428 if (!recovery_status) { 5429 err_rec->recovery_retries = 0; 5430 err_rec->recovery_state = ERR_RECOVERY_ST_NONE; 5431 dev_info(dev, "Adapter recovery successful\n"); 5432 goto reschedule_task; 5433 } else if (!lancer_chip(adapter) && err_rec->resched_delay) { 5434 /* BEx/SH recovery state machine */ 5435 if (adapter->pf_num == 0 && 5436 err_rec->recovery_state > ERR_RECOVERY_ST_DETECT) 5437 dev_err(&adapter->pdev->dev, 5438 "Adapter recovery in progress\n"); 5439 resched_delay = err_rec->resched_delay; 5440 goto reschedule_task; 5441 } else if (lancer_chip(adapter) && be_virtfn(adapter)) { 5442 /* For VFs, check if PF have allocated resources 5443 * every second. 5444 */ 5445 dev_err(dev, "Re-trying adapter recovery\n"); 5446 goto reschedule_task; 5447 } else if (lancer_chip(adapter) && err_rec->recovery_retries++ < 5448 ERR_RECOVERY_MAX_RETRY_COUNT) { 5449 /* In case of another error during recovery, it takes 30 sec 5450 * for adapter to come out of error. Retry error recovery after 5451 * this time interval. 5452 */ 5453 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n"); 5454 resched_delay = ERR_RECOVERY_RETRY_DELAY; 5455 goto reschedule_task; 5456 } else { 5457 dev_err(dev, "Adapter recovery failed\n"); 5458 dev_err(dev, "Please reboot server to recover\n"); 5459 } 5460 5461 return; 5462 5463 reschedule_task: 5464 be_schedule_err_detection(adapter, resched_delay); 5465 } 5466 5467 static void be_log_sfp_info(struct be_adapter *adapter) 5468 { 5469 int status; 5470 5471 status = be_cmd_query_sfp_info(adapter); 5472 if (!status) { 5473 dev_err(&adapter->pdev->dev, 5474 "Port %c: %s Vendor: %s part no: %s", 5475 adapter->port_name, 5476 be_misconfig_evt_port_state[adapter->phy_state], 5477 adapter->phy.vendor_name, 5478 adapter->phy.vendor_pn); 5479 } 5480 adapter->flags &= ~BE_FLAGS_PHY_MISCONFIGURED; 5481 } 5482 5483 static void be_worker(struct work_struct *work) 5484 { 5485 struct be_adapter *adapter = 5486 container_of(work, struct be_adapter, work.work); 5487 struct be_rx_obj *rxo; 5488 int i; 5489 5490 if (be_physfn(adapter) && 5491 MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0) 5492 be_cmd_get_die_temperature(adapter); 5493 5494 /* when interrupts are not yet enabled, just reap any pending 5495 * mcc completions 5496 */ 5497 if (!netif_running(adapter->netdev)) { 5498 local_bh_disable(); 5499 be_process_mcc(adapter); 5500 local_bh_enable(); 5501 goto reschedule; 5502 } 5503 5504 if (!adapter->stats_cmd_sent) { 5505 if (lancer_chip(adapter)) 5506 lancer_cmd_get_pport_stats(adapter, 5507 &adapter->stats_cmd); 5508 else 5509 be_cmd_get_stats(adapter, &adapter->stats_cmd); 5510 } 5511 5512 for_all_rx_queues(adapter, rxo, i) { 5513 /* Replenish RX-queues starved due to memory 5514 * allocation failures. 5515 */ 5516 if (rxo->rx_post_starved) 5517 be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST); 5518 } 5519 5520 /* EQ-delay update for Skyhawk is done while notifying EQ */ 5521 if (!skyhawk_chip(adapter)) 5522 be_eqd_update(adapter, false); 5523 5524 if (adapter->flags & BE_FLAGS_PHY_MISCONFIGURED) 5525 be_log_sfp_info(adapter); 5526 5527 reschedule: 5528 adapter->work_counter++; 5529 queue_delayed_work(be_wq, &adapter->work, msecs_to_jiffies(1000)); 5530 } 5531 5532 static void be_unmap_pci_bars(struct be_adapter *adapter) 5533 { 5534 if (adapter->csr) 5535 pci_iounmap(adapter->pdev, adapter->csr); 5536 if (adapter->db) 5537 pci_iounmap(adapter->pdev, adapter->db); 5538 if (adapter->pcicfg && adapter->pcicfg_mapped) 5539 pci_iounmap(adapter->pdev, adapter->pcicfg); 5540 } 5541 5542 static int db_bar(struct be_adapter *adapter) 5543 { 5544 if (lancer_chip(adapter) || be_virtfn(adapter)) 5545 return 0; 5546 else 5547 return 4; 5548 } 5549 5550 static int be_roce_map_pci_bars(struct be_adapter *adapter) 5551 { 5552 if (skyhawk_chip(adapter)) { 5553 adapter->roce_db.size = 4096; 5554 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev, 5555 db_bar(adapter)); 5556 adapter->roce_db.total_size = pci_resource_len(adapter->pdev, 5557 db_bar(adapter)); 5558 } 5559 return 0; 5560 } 5561 5562 static int be_map_pci_bars(struct be_adapter *adapter) 5563 { 5564 struct pci_dev *pdev = adapter->pdev; 5565 u8 __iomem *addr; 5566 u32 sli_intf; 5567 5568 pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf); 5569 adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >> 5570 SLI_INTF_FAMILY_SHIFT; 5571 adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0; 5572 5573 if (BEx_chip(adapter) && be_physfn(adapter)) { 5574 adapter->csr = pci_iomap(pdev, 2, 0); 5575 if (!adapter->csr) 5576 return -ENOMEM; 5577 } 5578 5579 addr = pci_iomap(pdev, db_bar(adapter), 0); 5580 if (!addr) 5581 goto pci_map_err; 5582 adapter->db = addr; 5583 5584 if (skyhawk_chip(adapter) || BEx_chip(adapter)) { 5585 if (be_physfn(adapter)) { 5586 /* PCICFG is the 2nd BAR in BE2 */ 5587 addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0); 5588 if (!addr) 5589 goto pci_map_err; 5590 adapter->pcicfg = addr; 5591 adapter->pcicfg_mapped = true; 5592 } else { 5593 adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET; 5594 adapter->pcicfg_mapped = false; 5595 } 5596 } 5597 5598 be_roce_map_pci_bars(adapter); 5599 return 0; 5600 5601 pci_map_err: 5602 dev_err(&pdev->dev, "Error in mapping PCI BARs\n"); 5603 be_unmap_pci_bars(adapter); 5604 return -ENOMEM; 5605 } 5606 5607 static void be_drv_cleanup(struct be_adapter *adapter) 5608 { 5609 struct be_dma_mem *mem = &adapter->mbox_mem_alloced; 5610 struct device *dev = &adapter->pdev->dev; 5611 5612 if (mem->va) 5613 dma_free_coherent(dev, mem->size, mem->va, mem->dma); 5614 5615 mem = &adapter->rx_filter; 5616 if (mem->va) 5617 dma_free_coherent(dev, mem->size, mem->va, mem->dma); 5618 5619 mem = &adapter->stats_cmd; 5620 if (mem->va) 5621 dma_free_coherent(dev, mem->size, mem->va, mem->dma); 5622 } 5623 5624 /* Allocate and initialize various fields in be_adapter struct */ 5625 static int be_drv_init(struct be_adapter *adapter) 5626 { 5627 struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced; 5628 struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem; 5629 struct be_dma_mem *rx_filter = &adapter->rx_filter; 5630 struct be_dma_mem *stats_cmd = &adapter->stats_cmd; 5631 struct device *dev = &adapter->pdev->dev; 5632 int status = 0; 5633 5634 mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16; 5635 mbox_mem_alloc->va = dma_alloc_coherent(dev, mbox_mem_alloc->size, 5636 &mbox_mem_alloc->dma, 5637 GFP_KERNEL); 5638 if (!mbox_mem_alloc->va) 5639 return -ENOMEM; 5640 5641 mbox_mem_align->size = sizeof(struct be_mcc_mailbox); 5642 mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16); 5643 mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16); 5644 5645 rx_filter->size = sizeof(struct be_cmd_req_rx_filter); 5646 rx_filter->va = dma_alloc_coherent(dev, rx_filter->size, 5647 &rx_filter->dma, GFP_KERNEL); 5648 if (!rx_filter->va) { 5649 status = -ENOMEM; 5650 goto free_mbox; 5651 } 5652 5653 if (lancer_chip(adapter)) 5654 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats); 5655 else if (BE2_chip(adapter)) 5656 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0); 5657 else if (BE3_chip(adapter)) 5658 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1); 5659 else 5660 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2); 5661 stats_cmd->va = dma_alloc_coherent(dev, stats_cmd->size, 5662 &stats_cmd->dma, GFP_KERNEL); 5663 if (!stats_cmd->va) { 5664 status = -ENOMEM; 5665 goto free_rx_filter; 5666 } 5667 5668 mutex_init(&adapter->mbox_lock); 5669 mutex_init(&adapter->rx_filter_lock); 5670 spin_lock_init(&adapter->mcc_lock); 5671 spin_lock_init(&adapter->mcc_cq_lock); 5672 init_completion(&adapter->et_cmd_compl); 5673 5674 pci_save_state(adapter->pdev); 5675 5676 INIT_DELAYED_WORK(&adapter->work, be_worker); 5677 5678 adapter->error_recovery.recovery_state = ERR_RECOVERY_ST_NONE; 5679 adapter->error_recovery.resched_delay = 0; 5680 INIT_DELAYED_WORK(&adapter->error_recovery.err_detection_work, 5681 be_err_detection_task); 5682 5683 adapter->rx_fc = true; 5684 adapter->tx_fc = true; 5685 5686 /* Must be a power of 2 or else MODULO will BUG_ON */ 5687 adapter->be_get_temp_freq = 64; 5688 5689 return 0; 5690 5691 free_rx_filter: 5692 dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma); 5693 free_mbox: 5694 dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va, 5695 mbox_mem_alloc->dma); 5696 return status; 5697 } 5698 5699 static void be_remove(struct pci_dev *pdev) 5700 { 5701 struct be_adapter *adapter = pci_get_drvdata(pdev); 5702 5703 if (!adapter) 5704 return; 5705 5706 be_roce_dev_remove(adapter); 5707 be_intr_set(adapter, false); 5708 5709 be_cancel_err_detection(adapter); 5710 5711 unregister_netdev(adapter->netdev); 5712 5713 be_clear(adapter); 5714 5715 if (!pci_vfs_assigned(adapter->pdev)) 5716 be_cmd_reset_function(adapter); 5717 5718 /* tell fw we're done with firing cmds */ 5719 be_cmd_fw_clean(adapter); 5720 5721 be_unmap_pci_bars(adapter); 5722 be_drv_cleanup(adapter); 5723 5724 pci_release_regions(pdev); 5725 pci_disable_device(pdev); 5726 5727 free_netdev(adapter->netdev); 5728 } 5729 5730 static ssize_t be_hwmon_show_temp(struct device *dev, 5731 struct device_attribute *dev_attr, 5732 char *buf) 5733 { 5734 struct be_adapter *adapter = dev_get_drvdata(dev); 5735 5736 /* Unit: millidegree Celsius */ 5737 if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP) 5738 return -EIO; 5739 else 5740 return sprintf(buf, "%u\n", 5741 adapter->hwmon_info.be_on_die_temp * 1000); 5742 } 5743 5744 static SENSOR_DEVICE_ATTR(temp1_input, 0444, 5745 be_hwmon_show_temp, NULL, 1); 5746 5747 static struct attribute *be_hwmon_attrs[] = { 5748 &sensor_dev_attr_temp1_input.dev_attr.attr, 5749 NULL 5750 }; 5751 5752 ATTRIBUTE_GROUPS(be_hwmon); 5753 5754 static char *mc_name(struct be_adapter *adapter) 5755 { 5756 char *str = ""; /* default */ 5757 5758 switch (adapter->mc_type) { 5759 case UMC: 5760 str = "UMC"; 5761 break; 5762 case FLEX10: 5763 str = "FLEX10"; 5764 break; 5765 case vNIC1: 5766 str = "vNIC-1"; 5767 break; 5768 case nPAR: 5769 str = "nPAR"; 5770 break; 5771 case UFP: 5772 str = "UFP"; 5773 break; 5774 case vNIC2: 5775 str = "vNIC-2"; 5776 break; 5777 default: 5778 str = ""; 5779 } 5780 5781 return str; 5782 } 5783 5784 static inline char *func_name(struct be_adapter *adapter) 5785 { 5786 return be_physfn(adapter) ? "PF" : "VF"; 5787 } 5788 5789 static inline char *nic_name(struct pci_dev *pdev) 5790 { 5791 switch (pdev->device) { 5792 case OC_DEVICE_ID1: 5793 return OC_NAME; 5794 case OC_DEVICE_ID2: 5795 return OC_NAME_BE; 5796 case OC_DEVICE_ID3: 5797 case OC_DEVICE_ID4: 5798 return OC_NAME_LANCER; 5799 case BE_DEVICE_ID2: 5800 return BE3_NAME; 5801 case OC_DEVICE_ID5: 5802 case OC_DEVICE_ID6: 5803 return OC_NAME_SH; 5804 default: 5805 return BE_NAME; 5806 } 5807 } 5808 5809 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id) 5810 { 5811 struct be_adapter *adapter; 5812 struct net_device *netdev; 5813 int status = 0; 5814 5815 status = pci_enable_device(pdev); 5816 if (status) 5817 goto do_none; 5818 5819 status = pci_request_regions(pdev, DRV_NAME); 5820 if (status) 5821 goto disable_dev; 5822 pci_set_master(pdev); 5823 5824 netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS); 5825 if (!netdev) { 5826 status = -ENOMEM; 5827 goto rel_reg; 5828 } 5829 adapter = netdev_priv(netdev); 5830 adapter->pdev = pdev; 5831 pci_set_drvdata(pdev, adapter); 5832 adapter->netdev = netdev; 5833 SET_NETDEV_DEV(netdev, &pdev->dev); 5834 5835 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 5836 if (status) { 5837 dev_err(&pdev->dev, "Could not set PCI DMA Mask\n"); 5838 goto free_netdev; 5839 } 5840 5841 status = be_map_pci_bars(adapter); 5842 if (status) 5843 goto free_netdev; 5844 5845 status = be_drv_init(adapter); 5846 if (status) 5847 goto unmap_bars; 5848 5849 status = be_setup(adapter); 5850 if (status) 5851 goto drv_cleanup; 5852 5853 be_netdev_init(netdev); 5854 status = register_netdev(netdev); 5855 if (status != 0) 5856 goto unsetup; 5857 5858 be_roce_dev_add(adapter); 5859 5860 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY); 5861 adapter->error_recovery.probe_time = jiffies; 5862 5863 /* On Die temperature not supported for VF. */ 5864 if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) { 5865 adapter->hwmon_info.hwmon_dev = 5866 devm_hwmon_device_register_with_groups(&pdev->dev, 5867 DRV_NAME, 5868 adapter, 5869 be_hwmon_groups); 5870 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP; 5871 } 5872 5873 dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev), 5874 func_name(adapter), mc_name(adapter), adapter->port_name); 5875 5876 return 0; 5877 5878 unsetup: 5879 be_clear(adapter); 5880 drv_cleanup: 5881 be_drv_cleanup(adapter); 5882 unmap_bars: 5883 be_unmap_pci_bars(adapter); 5884 free_netdev: 5885 free_netdev(netdev); 5886 rel_reg: 5887 pci_release_regions(pdev); 5888 disable_dev: 5889 pci_disable_device(pdev); 5890 do_none: 5891 dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev)); 5892 return status; 5893 } 5894 5895 static int __maybe_unused be_suspend(struct device *dev_d) 5896 { 5897 struct be_adapter *adapter = dev_get_drvdata(dev_d); 5898 5899 be_intr_set(adapter, false); 5900 be_cancel_err_detection(adapter); 5901 5902 be_cleanup(adapter); 5903 5904 return 0; 5905 } 5906 5907 static int __maybe_unused be_pci_resume(struct device *dev_d) 5908 { 5909 struct be_adapter *adapter = dev_get_drvdata(dev_d); 5910 int status = 0; 5911 5912 status = be_resume(adapter); 5913 if (status) 5914 return status; 5915 5916 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY); 5917 5918 return 0; 5919 } 5920 5921 /* 5922 * An FLR will stop BE from DMAing any data. 5923 */ 5924 static void be_shutdown(struct pci_dev *pdev) 5925 { 5926 struct be_adapter *adapter = pci_get_drvdata(pdev); 5927 5928 if (!adapter) 5929 return; 5930 5931 be_roce_dev_shutdown(adapter); 5932 cancel_delayed_work_sync(&adapter->work); 5933 be_cancel_err_detection(adapter); 5934 5935 netif_device_detach(adapter->netdev); 5936 5937 be_cmd_reset_function(adapter); 5938 5939 pci_disable_device(pdev); 5940 } 5941 5942 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev, 5943 pci_channel_state_t state) 5944 { 5945 struct be_adapter *adapter = pci_get_drvdata(pdev); 5946 5947 dev_err(&adapter->pdev->dev, "EEH error detected\n"); 5948 5949 be_roce_dev_remove(adapter); 5950 5951 if (!be_check_error(adapter, BE_ERROR_EEH)) { 5952 be_set_error(adapter, BE_ERROR_EEH); 5953 5954 be_cancel_err_detection(adapter); 5955 5956 be_cleanup(adapter); 5957 } 5958 5959 if (state == pci_channel_io_perm_failure) 5960 return PCI_ERS_RESULT_DISCONNECT; 5961 5962 pci_disable_device(pdev); 5963 5964 /* The error could cause the FW to trigger a flash debug dump. 5965 * Resetting the card while flash dump is in progress 5966 * can cause it not to recover; wait for it to finish. 5967 * Wait only for first function as it is needed only once per 5968 * adapter. 5969 */ 5970 if (pdev->devfn == 0) 5971 ssleep(30); 5972 5973 return PCI_ERS_RESULT_NEED_RESET; 5974 } 5975 5976 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev) 5977 { 5978 struct be_adapter *adapter = pci_get_drvdata(pdev); 5979 int status; 5980 5981 dev_info(&adapter->pdev->dev, "EEH reset\n"); 5982 5983 status = pci_enable_device(pdev); 5984 if (status) 5985 return PCI_ERS_RESULT_DISCONNECT; 5986 5987 pci_set_master(pdev); 5988 pci_restore_state(pdev); 5989 5990 /* Check if card is ok and fw is ready */ 5991 dev_info(&adapter->pdev->dev, 5992 "Waiting for FW to be ready after EEH reset\n"); 5993 status = be_fw_wait_ready(adapter); 5994 if (status) 5995 return PCI_ERS_RESULT_DISCONNECT; 5996 5997 be_clear_error(adapter, BE_CLEAR_ALL); 5998 return PCI_ERS_RESULT_RECOVERED; 5999 } 6000 6001 static void be_eeh_resume(struct pci_dev *pdev) 6002 { 6003 int status = 0; 6004 struct be_adapter *adapter = pci_get_drvdata(pdev); 6005 6006 dev_info(&adapter->pdev->dev, "EEH resume\n"); 6007 6008 pci_save_state(pdev); 6009 6010 status = be_resume(adapter); 6011 if (status) 6012 goto err; 6013 6014 be_roce_dev_add(adapter); 6015 6016 be_schedule_err_detection(adapter, ERR_DETECTION_DELAY); 6017 return; 6018 err: 6019 dev_err(&adapter->pdev->dev, "EEH resume failed\n"); 6020 } 6021 6022 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs) 6023 { 6024 struct be_adapter *adapter = pci_get_drvdata(pdev); 6025 struct be_resources vft_res = {0}; 6026 int status; 6027 6028 if (!num_vfs) 6029 be_vf_clear(adapter); 6030 6031 adapter->num_vfs = num_vfs; 6032 6033 if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) { 6034 dev_warn(&pdev->dev, 6035 "Cannot disable VFs while they are assigned\n"); 6036 return -EBUSY; 6037 } 6038 6039 /* When the HW is in SRIOV capable configuration, the PF-pool resources 6040 * are equally distributed across the max-number of VFs. The user may 6041 * request only a subset of the max-vfs to be enabled. 6042 * Based on num_vfs, redistribute the resources across num_vfs so that 6043 * each VF will have access to more number of resources. 6044 * This facility is not available in BE3 FW. 6045 * Also, this is done by FW in Lancer chip. 6046 */ 6047 if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) { 6048 be_calculate_vf_res(adapter, adapter->num_vfs, 6049 &vft_res); 6050 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 6051 adapter->num_vfs, &vft_res); 6052 if (status) 6053 dev_err(&pdev->dev, 6054 "Failed to optimize SR-IOV resources\n"); 6055 } 6056 6057 status = be_get_resources(adapter); 6058 if (status) 6059 return be_cmd_status(status); 6060 6061 /* Updating real_num_tx/rx_queues() requires rtnl_lock() */ 6062 rtnl_lock(); 6063 status = be_update_queues(adapter); 6064 rtnl_unlock(); 6065 if (status) 6066 return be_cmd_status(status); 6067 6068 if (adapter->num_vfs) 6069 status = be_vf_setup(adapter); 6070 6071 if (!status) 6072 return adapter->num_vfs; 6073 6074 return 0; 6075 } 6076 6077 static const struct pci_error_handlers be_eeh_handlers = { 6078 .error_detected = be_eeh_err_detected, 6079 .slot_reset = be_eeh_reset, 6080 .resume = be_eeh_resume, 6081 }; 6082 6083 static SIMPLE_DEV_PM_OPS(be_pci_pm_ops, be_suspend, be_pci_resume); 6084 6085 static struct pci_driver be_driver = { 6086 .name = DRV_NAME, 6087 .id_table = be_dev_ids, 6088 .probe = be_probe, 6089 .remove = be_remove, 6090 .driver.pm = &be_pci_pm_ops, 6091 .shutdown = be_shutdown, 6092 .sriov_configure = be_pci_sriov_configure, 6093 .err_handler = &be_eeh_handlers 6094 }; 6095 6096 static int __init be_init_module(void) 6097 { 6098 int status; 6099 6100 if (rx_frag_size != 8192 && rx_frag_size != 4096 && 6101 rx_frag_size != 2048) { 6102 printk(KERN_WARNING DRV_NAME 6103 " : Module param rx_frag_size must be 2048/4096/8192." 6104 " Using 2048\n"); 6105 rx_frag_size = 2048; 6106 } 6107 6108 if (num_vfs > 0) { 6109 pr_info(DRV_NAME " : Module param num_vfs is obsolete."); 6110 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n"); 6111 } 6112 6113 be_wq = create_singlethread_workqueue("be_wq"); 6114 if (!be_wq) { 6115 pr_warn(DRV_NAME "workqueue creation failed\n"); 6116 return -1; 6117 } 6118 6119 be_err_recovery_workq = 6120 create_singlethread_workqueue("be_err_recover"); 6121 if (!be_err_recovery_workq) 6122 pr_warn(DRV_NAME "Could not create error recovery workqueue\n"); 6123 6124 status = pci_register_driver(&be_driver); 6125 if (status) { 6126 destroy_workqueue(be_wq); 6127 be_destroy_err_recovery_workq(); 6128 } 6129 return status; 6130 } 6131 module_init(be_init_module); 6132 6133 static void __exit be_exit_module(void) 6134 { 6135 pci_unregister_driver(&be_driver); 6136 6137 be_destroy_err_recovery_workq(); 6138 6139 if (be_wq) 6140 destroy_workqueue(be_wq); 6141 } 6142 module_exit(be_exit_module); 6143